Open Access

What happened to real earnings in Egypt, 2008 to 2009?

IZA Journal of Labor & Development20132:10

DOI: 10.1186/2193-9020-2-10

Received: 25 June 2013

Accepted: 24 October 2013

Published: 25 November 2013

Abstract

Nominal earnings in Egypt did not respond to the increase in inflation between February 2008 and February 2009, resulting in a 12.3 (9) percent decline in average (median) real earnings among 25 to 60 years old workers. Changes in earnings differ significantly by groups: (i) those with higher initial earnings (reported and predicted) suffered the largest declines in earnings; (ii) men lost more than women but after controlling for initial earnings, women lost considerably more than men; (iii) those who were initially in agriculture also had greater declines despite rising food prices.

JEL classification codes

J62; J16; J42; J43; O15; O17

Keywords

Earnings mobility Egypt Inflation Gender Wage differentials Segmented labor market

1. Introduction

The returns to labor are critical for ensuring household economic well-being in Egypt, as elsewhere. As such, government leaders are concerned if they observe more labor lying dormant when unemployment rates rise or if they see the returns to labor fall when real earnings fall. In trying to better understand the workings and outcomes of the labor market and the returns to labor a multitude of studies have analyzed the Egyptian labor market. A number of these studies have examined earnings changes using cross-sectional data, illuminating important features of the Egyptian labor market (Arabsheibani 2000; Belhaj-Hassine 2012; Datt and Olmstead 2004; Said 2009 2012). However, comparisons of cross-sectional data from two points in time do not allow us to follow the same individual over time when trying to determine who is getting ahead and who is falling behind.

Recently, panel data has been collected that allows researchers to track the same individual over time. By following individuals over time, we can identify how an individual’s earnings level in the current year compares to his or her earnings in the previous period. We can also investigate how such earnings mobility experiences vary across individuals. This allows us to map the changes in earnings to the standard covariates, such as gender and education levels, and also to include new covariates such as the poverty level of the individual’s household and the earnings level of the individual in the initial period. Inclusion of these items can have an important impact on how we interpret the observed changes in the economy and resulting policy implications.

To our knowledge, this study represents the first analysis of earnings mobility in Egypt to incorporate initial earnings into analysis of earnings changes for both wage and non-wage workers1. We initially viewed this project as a study designed to establish baseline statistics for “typical” earnings mobility patterns in the Egyptian labor market.

However, given the significant inflation shock that occurred between interviews, we recognize that this may be a rather unique period in the labor market and should not be thought of as “typical”. Therefore we proffer this study as a window into the changes in the labor market in a time of severe stress on real wages due to an external shock of rising commodity prices.

Aggregate employment outcomes and nominal earnings or profits among prime-aged workers are shown to be fairly non-responsive despite the increased prices. Instead, inflation severely erodes the real earnings that individuals receive, implying an adjustment process that relies heavily on reduced real earnings. On average, a worker earns approximately 10 percent less than he or she did in the previous year.

The inability of workers to garner higher wages helped to keep inflation from continuing on an upward spiral2. The self-employed and employers also experienced declines in real profits suggesting that they either did not try or were not able to raise the prices of their goods and services in order to keep their real earnings constant or growing3.

The lack of labor market response to inflation in aggregate employment and nominal earnings outcomes should not be construed as a signal of rigidity of experiences at the individual level. As is typical in such studies, there is a large degree of heterogeneity in the earnings changes experienced by individuals. We identify key determinants associated with earnings mobility using a variety of descriptive regression techniques.

There is a strong association between initial earnings and subsequent earnings changes, with those initially earning more experiencing larger declines than their counterparts. As such changes can be caused by temporary fluctuations in earnings and/or measurement errors; this effect is taken with some caution. Nonetheless, we find these results generally robust to methods that address measurement error concerns, such as using predicted earnings in place of actual earnings. Additionally, other determinants of earnings changes that one would assume would be quite powerful predictors of earnings change—such as workers changing the sector of employment (agriculture, industry, or services)—have a limited ability to predict earnings mobility.

Changes in earnings also differ significantly by gender. On average, the earnings changes are far more negative for men than for women. However, this gender effect is completely reversed if one controls for initial earnings; i.e. for men and women who have the same initial earnings (and otherwise similar job characteristics), the earnings changes tend to be less negative for men than for women. In contrast, the average difference in log earnings is relatively similar across gender, with men having a slight advantage. This advantage for males is dramatically magnified if one controls for initial log earnings.

The rest of the paper is organized as follows. Section 2 first describes the context of our study, and explains the context of the Egyptian economy during this time, including a review of the rise in inflation. Section 3 depicts the panel data used in this study. Section 4 portrays the empirical approach. Section 5 introduces the results. Section 6 provides a discussion of these results and presents the relevance to policy and future research.

2. Background discussion of the Egyptian economy

The World Bank study, “Poverty in Egypt 2008–2009: Withstanding the global economic crisis,” includes a substantive review of the Egyptian economy leading up to 2008 (World Bank 2011). In this section, we review some of its critical findings relevant to our study.

From 2004 to mid-2008, Egypt experienced a period of steady economic growth with an average real GDP annual growth of 6.4 percent. This growth period was followed by a series of economic shocks starting in the second half of 2008 and continuing for most of 2009, namely:
  1. (i)

    the slowdown in economic growth, which followed the global downturn in the world economy;

     
  2. (ii)

    a simultaneous and related fall in remittances; and

     
  3. (iii)

    a sudden acceleration in prices, mostly driven by exogenous factors such as the commodities price shock.

     
The economic developments over the period 2004–2009 are summarized in Figure 1. Note that the table begins in the last quarter of 2004 using the calendar year (second quarter of the fiscal year). For reference, the panel data analyzed in this study were collected in the first quarter of 2008 and the first quarter of 2009. They are denoted by the vertical strips at the third quarter of fiscal year 2008 and 2009, respectively.
Figure 1

Quarterly GDP growth, inflation rate and remittances in Egypt, 2004–2009. Reprinted from The World Bank (2011). Data from CAPMAS and Ministry of Economic Development. Note: The quarterly GDP per capita data were annualized before determining the percent increase or decrease in the quarter-on-quarter index. CPI data were not annualized.

At first glance, Egypt weathered the global crisis relatively well. Despite an abrupt economic slowdown, from seven percent growth in 2008 to 4.7 percent in 2009, growth remained positive. However, employment growth stalled in absolute terms (employment was flat during 2008 and the first quarter of 2009), and unemployment started to rise again (to 9.4 percent in the first quarter of 2009). According to the Labor Force Sample Survey (LFSS) conducted by the Central Agency for Public Mobilization and Statistics (CAPMAS), there was a seven percent reduction in employment in manufacturing in the first quarter of 2009 (compared to the first quarter of 2008), and a 15 percent drop in the restaurants and hotel sectors. Meaning that, both sectors are most exposed to global trends and thus more susceptible to global shocks.

These changes came during a period of accelerating inflation, which peaked at 20 percent (annual average) in the summer of 2008. After peaking to an unprecedented level in August 2008 (23.7 percent), inflation remained exceptionally high for two quarters, only to stabilize at 9.9 percent between May and July-09. Inflation was almost completely imported, as a result of the sharp acceleration in global commodities prices (mostly fuel and food, including fruits and vegetables). Some components of the inflation basket, such as food and non-alcoholic beverages, remained high (up by 13.4 percent y-to-y in July-09) even after and despite the steep fall in international prices since mid-2008.

By the last quarter of 2009, the Egyptian economy showed signs of rebounding, and since then has been growing at five percent and higher each quarter. The employment losses were soon recovered (at least in absolute figures). Thus, our earnings mobility analysis may overestimate some of the long term decline in earnings that individuals faced from this period.

3. Data

The data used in this analysis come from the Household Income, Expenditure and Consumption Panel Survey (HIECPS), which is a subcomponent of the Household Income, Expenditure and Consumption Survey (HIECS). This is a nationally representative panel survey. However, as there were relatively few observations from the frontier governorates, we exclude the border region from our analysis.

Households were interviewed in February 2008 and those same households were interviewed again in February 2009. Information gathered included individual labor market outcomes, wages and profits from enterprises run by household members, as well as standard demographic information such as age, gender, and education level of all household members.

Earnings were constructed at the individual level. Earnings from household enterprises were split among all household members engaged in the enterprise, including those typically referred to as “unpaid workers4”. Unless clearly denoted as nominal earnings, all earnings are deflated over both time and space and represent the purchasing power of the 2005 Egyptian pound (EGP) spent in the metropolitan area.

The analysis is limited to those who were 25 to 58 years old in 2008, unless explicitly stated. We are most interested in the changes in earnings of those actively engaged in the labor market and do not want large earnings changes by those retiring or moving to part time jobs in a pseudo-retirement phase to cloud our analysis. Additionally, large gains from recent entrants (or entrants into full time work) can disrupt our analysis. This is not to say that we are not interested in the important issues of youth (15–24 years) unemployment or earnings. We simply feel they should be analyzed separately. We will discuss how the inclusion of these younger workers affects some of our key results below.

Additionally, we limit earnings mobility analysis to those who are employed in both periods5. We use 3,481 observations of dual employed individuals with valid data.

One concern of such panel studies is that not every person interviewed in 2008 can be found again (or is willing to be re-interviewed) in 2009. This can have a particularly pernicious effect on our analysis if those who were not re-interviewed tend to have (extremely) positive or (extremely) negative earnings changes compared to those who are similar to them but were re-interviewed. There is some empirical evidence that such movers, particularly those that move farther away, tend to have more positive earnings changes than those left behind (Beegle et al. 2011). However, the existing evidence is quite limited and not from Egypt. Additionally, there are some common sense theoretical arguments that suggest that such missing individuals could be worse off. This question remains open as we do not have enough evidence to answer it in one direction or the other.

The fact that attrition rates in this panel are relatively low (just three percent of 2008 households are not re-interviewed in 2009) is therefore reassuring. As is typical in such studies, the individual attrition rate, 9.1 percent, is higher than the household attrition rate. Among 25 to 58 year olds, the individual attrition rate is 8.7 percent.

Overall, attrition appears to be driven more by who a person is and general life-cycle effects than by an individual’s productive characteristics or outcomes. The first column of Table 1 shows the results of a linear probability model (LPM) predicting individual attrition. The dependent variable is 1 if the individual is not re-interviewed and 0 if the individual is re-interviewed. Those who were not married, were either very young or very old, were male and lived in the Metropolitan region were more likely to be missing at the 2009 interview. Meanwhile, joint significance tests on education level, initial sector of work, and relationship with employer categories were not found to have a significant effect at the ten percent level. An exception to this general trend is the fact that those who had higher initial earnings were found to be more likely to attrite. This was statistically significant at the five percent level, with change of one standard deviation in real earnings in 2008 resulting in a change in the predicted probability of attrition of 0.009 percentage points, ceteris paribus. A probit model was also run (see rightmost column of Table 1). The results of the probit model generally corroborate the LPM findings. One potential exception is that individuals working in the manufacturing and construction and trades and services sectors are less likely to attrite than those working in agriculture6.
Table 1

Determinants of attrition

 

Dep. Var. (Attrition = 1)

 

Linear

Probit

 

Prob. Model

dF/dx

Metropolitan (omitted)

  

Lower Urban

−0.061***

−0.039***

Lower Rural

−0.056***

−0.043***

Upper Urban

−0.045**

−0.029**

Upper Rural

−0.063***

−0.046***

Male

0.041***

0.030***

Working Housewife

−0.024

−0.029*

Married

−0.091***

−0.059***

Age

−0.022***

−0.017***

Age-squared

0.000***

0.0001***

Illiterate/no formal schooling (omitted)

  

Basic education/literacy

0.008

0.010

Completed Primary

0.004

0.006

Secondary education

−0.005

−0.003

University degree

0.025

0.022

wage worker public (omitted)

  

wage worker private

0.018

0.015

employer

−0.019

−0.015

self employed

−0.002

−0.001

unpaid worker

0.007

0.002

Agriculture (omitted)

  

Manu/Const

−0.032*

−0.025**

Trade/Services

−0.021

−0.022*

Log per capita consumption

0.008

0.009

Earnings in 2008

−0.000**

0.000

Constant

0.635***

 

n

4160

4160

R-squared / Pseudo R-squared

0.076

0.112

Notes: legend: * p < 0.10; ** p < 0.05; *** p < 0.01. Probit dF/dx evaluated at the mean of all other variables. Dummy variables show difference from 0 to 1.

Source: HIECPS, Feb 2008 to Feb 2009 (CAPMAS 2009).

4. Empirical approach

We identify the typical change in earnings experienced by different types of Egyptian workers using empirical approaches that are common in the earnings and income mobility literature. We start by reporting mean and median earnings changes and move on to control for an increasing set of covariates. Techniques are similar to some of those used in Cichello et al. (2005), Fields et al. (2003a/2003b), and Fields and Sanchez-Puerta (2010). Fields (2008) provides a literature review of such micro-mobility studies and provides a description of methods. There is a particular emphasis on controlling for initial earnings.

The goal of this empirical approach is to identify who has experienced the most severe declines in earnings, who has been able to avoid such severe declines in earnings, and which variables are consistently associated with the changes in earnings experienced by Egyptian workers. While all of the analysis, even the multivariate regression analysis, is descriptive rather than causal, one can confidently identify which groups of workers have been most or least adversely affected7. One can also piece together strong circumstantial evidence regarding some of the major drivers of earnings change. Theories of what caused the changes in earnings should, at a minimum, be consistent with these descriptive results.

Our analysis will use both the change in earnings and the change in log earnings as the outcome variables of interest. The former is self-explanatory while the latter conforms roughly to the percentage change in earnings. Given that utility functions are concave, determining that earnings have fallen more on average for those who started with higher earnings may not necessarily suggest a lower welfare loss8. Also, in the context of an inflation shock, one may not be overly surprised if those with higher initial earnings have greater earnings declines. However, if the log earnings falls more for those with higher initial earnings, this goes beyond just the changes due to inflation.

We assess the change in earnings across a variety of characteristics. These include demographic variables (gender, age, marital status), location/region (urban and rural), and education level. We take these variables as exogenous variables that individuals bring to the labor market9. We also include a number of 2008 outcomes which may well be co-determined with the earnings change process. These include an individual’s sector of employment (agriculture, industry or construction, trade or services); relationship with employer (public wage worker, private wage worker, employer, self-employed, unpaid worker); and formal or informal worker status, with formal workers defined as those covered by the social insurance system. We also include whether one lives in a poor household, defined as having consumption levels below the household specific poverty line10. In our multivariate analysis, these variables will be right hand side determinants of the change in log earnings.

Additionally, the relationship between the change in earnings and initial earnings is explored using flexible estimation approaches. The non-parametric lowess estimator estimates a smoothed change in log earnings value, yi s, for each observation by running n observation specific regressions of the change in log earnings (yi) on initial log earnings (xi) and calculating the predicted value of yi. Weights for each observation in the regression are determined based on the distance from the xi value of the current observation. More emphasis is placed on observations that are close to the current observation, even going so far as to give a zero weight to observations outside the bandwidth. The lowess estimator in STATA uses the tri-cube weight of Cleveland (1979) and we apply a bandwidth of 0.8 throughout these figures11.

Measurement error can be particularly problematic when determining the change in earnings to initial earnings. For example, assume the relationship between these variables is linear and we would like to run a simple regression of change in reported earnings Δyr i, on initial reported earnings yr 08,i.
Δ y r i = β 0 + β 1 y r 08 , i + ϵ i ,
(1)
This can be re-written incorporating the potential for measurement error, signified by μ. We use yk to represent the true level of earnings in year k.
Δ y i + µ 09 , i µ 08 , i = β 0 + β 1 y 08 , i + µ 08 , i + ϵ i ,
(2)

Even if measurement error is assumed to be mean zero, independent across individuals, independent across time for a given individual and independent of an individual’s earnings level, the problems are more than the standard attenuation concerns that typically accompany a classical measurement error. The realized measurement error in year 2008 turns up on both the left hand and right hand sides of this equation causing a spurious negative correlation. To the extent that the measurement error is not perfectly auto-correlated, this can lead to a problem identical in nature to Galton’s Fallacy. A negative relationship between initial earnings and the change in earnings might simply be due to random reporting errors.

Running the above regression will not give us the desired regression coefficient,
β ^ * = Cov Δy , y 08 Var y 08
(3)
Instead, as shown in a more general form in Fields et al. (2003a), we obtain:
β ^ = β ^ * Var y 08 Var y 08 + Var μ 08 Var μ 08 Var y 08 + Var μ 08
(4)

The first term identifies the standard attenuation bias issue while the second term accounts for the spurious negative correlation caused by the measurement error being on both the right and left side of the equation.

A simple example will illustrate why addressing this issue is of fundamental importance to our analysis. Suppose β ^ * = 0.50 , implying that for each additional pound of earnings in 2008, on average, the individuals experiences a change in earnings that is a half a pound more (i.e. less negative). Also, suppose that the variance in measurement error is exactly equal to the variance in initial earnings. In this case, the attenuation bias would shrink the estimate from 0.5 to 0.25 and the second term would take away another 0.50. In other words, we would observe a coefficient of −0.25 in our analysis. Our story would be one of converging earnings, but the reality would be significant divergence of earnings in this sample. This example, while extreme, shows how powerful the impact of these biases can be on our results. It also highlights that the extent of the problem is driven in part by our assumption of how large the variance of the measurement error is relative to the variance of the initial earnings. For example, if we assumed the variance of μ was ten percent of the variance of initial earnings, the coefficient would observe would be 0.35 rather than 0.5012.

In order to assess the robustness of our results given this potential problem, two approaches will be applied. First, initial earnings will be substituted by predicted initial earnings on the right hand side of equation (1) in the simple regression above (as well as in other methods). This process, i.e., the use of predicted values of initial values, should rid the right hand side of equation (1) of the spurious measurement error13. Of course, if our ability to predict initial earnings is weak, we may lose more than we gain from this approach. It should be noted that the key variables used to predict initial earnings are gender, age, education, place of residence, relationship with employer, sector of work, and log per capita consumption in the household.

The prediction regression results are shown in Table 2. When predicting initial earnings, the R-squared is 0.32. When predicting initial log earnings, the R-squared is .612. Thus, there is a relatively strong ability to predict initial earnings in levels and logs. The coefficient estimates are entirely reasonable; they are what we would expect in Egypt. Ceteris paribus, men have a strong initial advantage, earnings are concave with respect to age (turning point at roughly 45 years of age), those with higher education are earning considerably more than those with less education, wage workers outperform the self-employed and unpaid workers, but not employers, and agricultural workers earn less than others.
Table 2

Regressions used to predict 2008 earnings and log earnings

 

2008 earnings

2008 log earnings

Male

4517***

0.580***

Age

723***

0.090***

Age-squared

−8***

−0.001***

Working Housewife

603

−0.207***

Married

2873***

0.366***

Illiterate/no formal schooling (omitted)

  

Basic education/literacy

414

0.094***

Completed Primary

1700***

0.155***

Secondary education

636*

0.115***

University degree

1406***

0.194***

Metropolitan (omitted)

  

Lower Urban

−1035**

−0.007

Lower Rural

−1495***

−0.049*

Upper Urban

−563

0.033

Upper Rural

−439

−0.021

wage worker public (omitted)

  

wage worker private

−356

−0.032

employer

3724***

0.388***

self employed

152

−0.114***

unpaid worker

−882

−0.509***

Agriculture (omitted)

  

Manu/Const

1021**

0.148***

Trade/Services

775*

0.126***

log per capita consumption (08)

5748***

0.520***

Constant

−60681***

1.682***

n

3481

3481

R-squared

0.32

0.612

Notes: legend: * p < 0.10; ** p < 0.05; *** p < 0.01.

Source: HIECPS, Feb 2008 to Feb 2009 (CAPMAS 2009).

We will also assess whether such a classical measurement error can explain the results we observe in this dataset using a second approach. We assess the amount of measurement error that must be present, defined in terms of the variance, for measurement error to explain our observed regression coefficient if the true β coefficient is positive; i.e. if those higher initial earners experienced greater earnings gains. This is similar to a subset of simulations found in Fields et al. (Fields et al. 2003a). Additionally, in the annex, we use a simple simulation to demonstrate what our lowess graphs would look like given certain assumed levels of classical measurement error.

Under these assumptions, we consider it implausible that the negative coefficient on initial earnings from the simple regression was driven solely by classical measurement error. In order for this to be true, the variance of measurement error must be at least 1.56 times the variance of the true earnings level.

5. Findings

5.1 Transitions into and out of employment

Employment participation held steady and even increased slightly (68.2 to 69.6 percent), between February 2008 and February 2009 among panel members who were 25 to 58 years old in 2008. Thus, this is not a period where job losses were a dominant feature in the labor market for prime-aged workers14.

Table 3 reveals that the aggregate participation rate masks strikingly different participation rates by gender, 94.5 percent for men versus 44.4 percent for women in 2008. The difference is largely due to many more women remaining outside the labor force.
Table 3

Labor force status of 25 to 58 year olds in 2008 and 2009, by gender

 

All

Male

Female

 

2008

2009

2008

2009

2008

2009

Employed

68.2

69.6

94.5

94.9

44.4

46.7

Unemployed

1.9

1.6

1.7

1.5

2.1

1.7

Out of Labor Force

29.9

28.8

3.8

3.6

53.6

51.6

Notes: n = 5702.

Source: HIECPS, Feb 2008 to Feb 2009 (CAPMAS 2009).

Given the decline in real earnings between 2008 and 2009 (which is discussed in much more detail below), one might expect households to offset the erosion of real earnings by increasing the number of earners in the households. Since most men are already participating, this would require an influx of women into the labor market.

The “added worker effect” is discussed in the labor literature primarily with regard to married women entering the labor market in response to unemployment of the male (Lundberg 1985; Serneels 2002). The empirical evidence has been relatively weak for any large scale effects on employment rates in other countries, and often weak for even desiring more employment (Serneels 2002). Our situation differs from the often cited situation in that the male is not unemployed and therefore not necessarily “freely” available to substitute for the wife in the production of goods and services at the home. Additionally, the wife’s real wage has also been reduced. Thus, we are not necessarily expecting a large change in female participation in response to the reduced earnings power of the male.

We find that the employment participation rates increase slightly for both males and females in 2009, to 94.9 percent and 46.7 percent respectively. These numbers do not suggest that households are adding workers as a major strategy for coping with inflation. Nor does the data suggest that firms fired large numbers of workers as a result of inflation15. These finding are consistent with findings of Roushdy and Gadallah (2011) which finds relatively little change in employment outcomes despite the economic downturn and large real earnings declines in 2009.

Table 4 takes advantage of the longitudinal nature of the data to allow us to see the flows into and out of each labor market status. The tables show the percentage of men (women) in each transition possibility, with the sum of all cells equaling 100 percent.
Table 4

Labor force transitions of 25 to 58 year olds in 2008 and 2009, by gender

Male

   
  

2009

 

Status in 2008

Employed

Unemp

OLF

Employed

92.4

0.6

1.5

Unemployed

1.0

0.5

0.2

Out of Labor Force

1.6

0.3

1.9

Female

   
  

2009

 

Status in 2008

Employed

Unemp

OLF

Employed

36.2

0.4

7.8

Unemployed

0.4

0.7

1.0

Out of Labor Force

10.1

0.7

42.7

Notes: n = 2702 and 2995, respectively.

Source: HIECPS, Feb 2008 to Feb 2009 (CAPMAS 2009).

Most interesting is the large flow of women moving into and out of employment from the ‘out of the labor force’ category. Approximately one in six of women employed in 2008 are classified as out of the labor force in 2009, with roughly similar percentages of those out of the labor force in 2008 moving into employment in 2009. It is possible that some of this apparent flow is due to mis-measurement as women may not be declared employed in one period. However, it suggests that there are a large number of women who are capable of entering the labor force, at least for some time, if the household desires additional earnings. The story is similar if the data is restricted to only married women (not shown). Thus, there appears to be some scope culturally for many houses to add women workers to the labor force if needed. Finally, there is actually a slight decline in the unemployment rate for women, from 4.5 percent in 2008 to 3.5 percent in 2009.

While the bulk of men are employed in both periods, there is some evidence that the limited set of men who are unemployed may remain unemployed for some time. Thirty percent of those men who were unemployed in 2008 were unemployed again when interviewed in 2009 (as compared to just 0.7 percent who were employed at the time of the interview in 2008). The unemployment issue is similar for females as 31 percent of females who were unemployed in 2008 were unemployed again in 2009. However, in Egypt, unemployment is generally considered a luxury, where highly educated workers wait for a high paying job. For example, 85 percent of those unemployed in 2008 have secondary or university education, as compared to 45 percent of the entire prime-age working population.

5.2 A period of declining real earnings

Restricting our attention to the dual employed, the mean earnings change was a loss of 950 EGP, with the median change being a loss of 446 EGP. As will be detailed below, this decline in mean and median earnings was experienced by almost every type of worker. This is consistent with the erosion in real pay found by Said using the quarterly Egypt Labor Force Survey (ELFS) from 2007–2009 (Said 2012). The decline in real earnings is the dominant story of the labor market during this time period.

This decline in real earnings appears directly related to the increase in inflation. The mean (median) change in nominal earnings is just 107(330) EGP compared to a mean (median) of 9,858 (7,890) EGP in 2008 nominal earnings. Alternatively, if we apply the 2009 deflators to the 2008 nominal earnings, we project the average earnings fall from 7,755 EGP in 2008 to 6,727 EGP in 2009. In fact, the average real earnings in 2009 are 6,804 suggesting that workers earnings basically stagnated while inflation took away its buying power.

The improvement when comparing the median earnings level in each year is considerably better. The 6,237 EGP median earnings in 2008 would equate to 5,424 EGP using the 2009 deflators while the real median earnings in 2009 was 5,832EGP. Thus, comparing median earnings levels across distributions, workers were able to get back approximately 50 percent of the loss in earnings due to inflation. However, this seems to be an unusual point in the distribution of earnings. Corresponding values at the 25th and 75th percentiles are 16 percent and 23 percent. Figure 2 depicts the percentage of earnings lost to inflation that earners were able to recover due to increased nominal earnings for each percentile. At the top and bottom of the distributions, earnings appear to fall even in nominal terms. Overall, nominal earnings don’t seem to increase much in response to the rapid rise in inflation, resulting in a significant loss of earnings power.
Figure 2

Percent of real earnings lost to inflation that was regained due to increased nominal pay, Egypt 2008 to 2009. Source: Authors’ compilation from HIECPS (Feb 2008 to Feb 2009 (CAPMAS 2009)). Notes: This figure shows the change in nominal earnings for a given percentile in the earnings distribution as a percentage of the loss in real earnings that would have been experienced due to increased inflation if nominal earnings were the same in 2008 and 2009. The first percentile value (−161) has not been shown as it distorts the scale. Negative percentages imply that nominal earnings have declined between 2008 and 2009 for that percentile of the earnings distribution.

This result is in keeping with other findings from the Egyptian labor market. For example, Datt and Olmsted suggest that the rise and fall of real wages in Egypt between the mid-1970s and the early 1990s “masks a complex dynamic process by which nominal wages adjust in response to changes in food prices” (Datt and Olmsted 2004). The authors show a slow adjustment of nominal wages, with “a significant negative initial impact of rising food prices on real wages, though wages do catch up in the long run” (ibid).

Figure 3 depicts the real earnings changes using a density function of the change in log earnings for all but the top and bottom 2.5 percent of changes. The distribution is centered below zero, with a median change in log earnings of negative 0.09 and a mean of negative 0.123.
Figure 3

Changes in log earnings in Egypt, 2008 to 2009. Source: HIECPS (Feb 2008 to Feb 2009 (CAPMAS 2009)). Notes: Density function excludes top and bottom 2.5 percent of changes. The vertical line depicts the median change in log earnings.

Despite the substantial decline in earnings for measures of central tendency, the picture is indeed more complex than one where every individual earns 12 percent less than the year before: 41 percent of individuals experienced a gain in earnings over the time period; many individuals experienced positive and negative changes that were quite sizeable relative to their initial earnings; 31 percent of the changes in the density function above are greater than 0.5 or less than −0.5. While we would expect some differential across years due solely to the measurement error, we would expect that differentials of 50 percent in reported earnings would be extremely rare if an individual had the same earnings in both years16.

As this is the first earnings mobility study in Egypt, we cannot say if this is more or less volatility in earnings than is the norm. What is clear is the following: while the aggregate employment numbers suggest a stagnant situation overall—with little change in employment status and average and median nominal wages—individuals are experiencing substantial fluctuations in their earnings. The rest of this paper seeks to explore which characteristics are associated with changes in earnings at the individual level.

5.3 The influence of initial earnings

There is a strong negative relationship between initial earnings and the ensuing earnings change. Using simple regression, Table 5 shows a statistically significant relationship. The coefficient estimate implies that the predicted earnings change is 0.61 EGP less for every additional Egyptian pound of reported initial earnings.
Table 5

Simple regression of change in earnings on initial earnings

Regression

Coeff.

st. error

p-value

Δy on y

−0.61

0.097

0.000

Δy on predicted y

−0.29

0.052

0.000

Δln(y) on ln(y)

−0.23

0.017

0.000

Δln(y) on predicted ln(y)

−0.04

0.026

0.131

Notes: Standard errors account for strata and clusters in sample survey design. Standard errors are bootstrapped when using predicted earnings or predicted log earnings.

Source: HIECPS, Feb 2008 to Feb 2009 (CAPMAS 2009).

Table 5 shows that the negative relationship between earnings change and initial earnings holds up even if one replaces reported initial earnings with predicted initial earnings, although the coefficient falls to −0.29. Predicted initial earnings should not be correlated with the measurement error term under standard classical measurement error assumptions, although the random error in the predicted value of initial earnings may result in attenuation bias.

We consider it implausible that the negative coefficient on initial reported earnings from the simple regression was driven solely by classical measurement error. In order for this to be true, under the assumptions outlined above and the resulting equation (4), the variance of measurement error must be at least 1.54 times the variance of the true earnings level.

The analysis is repeated using log earnings. For every one percent increase in initial earnings, the predicted change in earnings decreases by approximately 0.23 percent. This result is statistically significant at the 1 percent level. In this case, again using equation (4), the variance of measurement error need only be 30 percent of the variance of the true earnings level for it to entirely explain the observed negative coefficient17. Additionally, we find that the predicted change in earnings decreases by 0.04 percent for every 1 percent increase in predicted initial log earnings. However, the p-value for this coefficient lies slightly outside the ten percent range for statistical significance. Thus, we are not confident that the true coefficient is negative.

This analysis assumes a linear relationship between initial earnings and earnings change. Next we allow more flexible approaches of establishing this relationship. Table 6 presents the mean and median change in earnings for each quintile of initial earnings. It also documents the percent of individuals in each group that experienced positive changes in earnings.
Table 6

Change in earnings by initial earnings quintile

Initial reported earnings quintile

Change in earnings

Change in log earnings

 

Mean

Median

% positive

Mean

Median

% positive

1st (Lowest)

567

122

56%

0.086

0.071

56%

2nd

502

142

54%

−0.045

0.035

54%

3rd

119

−245

45%

−0.080

−0.040

45%

4th

−914

−1,148

31%

−0.200

−0.143

31%

5th

−5,026

−3,825

19%

−0.375

−0.294

19%

Initial Predicted (Log) Earnings Quintile

 

Mean

Median

% positive

Mean

Median

% positive

1st (Lowest)

20

−45

48%

−0.124

−0.073

45%

2nd

−290

−295

44%

−0.103

−0.039

47%

3rd

−417

−306

45%

−0.062

−0.055

45%

4th

−867

−939

36%

−0.133

−0.111

37%

5th

−3,199

−1,880

32%

−0.193

−0.161

32%

TOTAL

−950

−446

41%

−0.123

−0.090

41%

The results clearly demonstrate a downward trend in the change in earnings as initial earnings quintile increases each step. The change between the first and second quintiles is rather limited, but other changes are quite large. This result is fairly consistent when comparing the change in earnings across predicted earnings quintiles. There is a clear downward trend in mean, median and percent positive earnings change across initial predicted earning quintile, except some bunching or minor reversal at the second and third quintiles.

When comparing the change in log earnings across initial log earnings quintiles, the results demonstrate a clear downward trend by initial earnings quintile. When comparing the change in log earnings across initial predicted log earnings, however, the data shows some potential for an inverted U shape, where those in the middle earnings category were able to buffer their losses (in percentage terms) as compared to those in the lowest earnings quintiles. Overall, however, the typical decline is far worse for those at the highest earnings levels as compared to those initially at the lowest earnings levels.

The severe decline in earnings experienced by those in the highest quintile is consistent across all four models. Just 19 percent of those in the highest initial earnings category experienced positive earnings changes. The mean decline in earnings for this group was 5,026 EGP as compared to the average of 950 EGP across the full sample. The median decline in earnings was 3,825 EGP and the median difference in log earnings was −0.29.

Figure 4 presents a more flexible method for assessing the relation between changes in log earnings and initial log earnings using a locally weighted regression. The vertical lines denote the different quintiles of initial log earnings. These results confirm the downward sloping relationship between the change in log earnings and initial log earnings, with bunching at the second and third quintile.
Figure 4

Change in log earnings based on initial log earnings, Egypt 2008–2009. Source: HIECPS (Feb 2008 to Feb 2009 (CAPMAS 2009)). Notes: Locally weighted regression was used to determine the predicted change in log earnings based on log real earnings in 2008. The top and bottom one percent of changes in log earnings are excluded. Vertical lines denote the 20th, 40th, 60th and 80th percentiles for 2008 log real earnings.

This result is consistent with previous examinations of real earnings that show declining inequality in periods when real earnings are falling in Egypt (Said 2009).

5.4 The role of characteristics besides initial earnings

Table 7 presents the mean and median earnings change and the percent of positive earnings changes for a series of additional characteristics that may be useful for either understanding more about the underlying process that drove the earnings changes during this time period and/or for understanding who was most adversely affected for the purpose of targeting assistance. This information is presented in columns (3) through (5), respectively. Columns (6) and (7) present the mean and median change in log earnings. Columns (1) and (2) are provided solely to assist the reader in understanding the composition of the labor force and relative earnings standings of each type of worker in the initial period.
Table 7

Change in earnings and log earnings by individual characteristics, Egypt 2008 to 2009

2008

 

Change in earnings

Change in ln earnings

(1)

(2)

 

(3)

(4)

(5)

(6)

(7)

Percent of panel workers

Median earnings

Individual characteristics

Mean

Median

Percent positive

Mean

Median

  

TOTAL

     

100%

6,237

 

-950

-446

41%

-0.12

-0.09

  

Gender

     

30%

2,964

    Female

-391

-206

44%

-0.16

-0.08

70%

7,627

    Male

-1,189

-628

40%

-0.11

-0.10

  

Age

     

31%

4,648

    25-34

-459

-317

43%

-0.09

-0.08

34%

6,778

    35-44

-803

-421

42%

-0.10

-0.07

35%

7,478

    45-58

-1,518

-639

38%

-0.18

-0.12

  

Education

     

30%

4,047

    Illiterate/no formal schooling

-781

-471

39%

-0.19

-0.14

12%

7,010

    Basic education/literacy

-1,534

-815

34%

-0.17

-0.16

10%

7,072

    Completed Primary

-2,070

-699

41%

-0.14

-0.10

32%

6,503

    Secondary education

-610

-287

43%

-0.07

-0.06

16%

8,490

    University degree

-850

-349

46%

-0.04

-0.05

  

Urban

     

62%

5,316

    Rural

-648

-404

41%

-0.14

-0.10

38%

7,783

    Urban

-1,437

-516

42%

-0.10

-0.08

  

Region

     

17%

8,272

    Metropolitan

-2,318

-817

39%

-0.13

-0.14

11%

8,481

    Lower Urban

-413

-214

46%

-0.03

-0.03

38%

5,421

    Lower Rural

-534

-287

43%

-0.11

-0.08

10%

6,859

    Upper Urban

-1,053

-620

42%

-0.14

-0.08

24%

5,068

    Upper Rural

-827

-522

37%

-0.18

-0.14

  

Type of worker

     

33%

7,400

    wage worker public

-633

-117

47%

-0.03

-0.02

25%

5,914

    wage worker private

-628

-504

41%

-0.10

-0.10

15%

9,324

    employer

-2,809

-1,803

28%

-0.30

-0.21

16%

4,054

    self employed

-1,077

-522

38%

-0.25

-0.16

10%

1,925

    unpaid worker

214

-70

47%

-0.01

-0.04

  

Informal worker

     

53%

4,603

    Informal

-838

-471

40%

-0.16

-0.13

47%

7,827

    Formal

-1,074

-395

42%

-0.08

-0.06

  

Industry

     

31%

3,411

    Agriculture

-829

-427

39%

-0.20

-0.17

19%

7,010

    Manu/Const

-1,200

-645

37%

-0.12

-0.10

51%

7,151

    Trade/Services

-931

-361

44%

-0.08

-0.06

  

Poor household in 2008

     

84%

6,893

    Non-poor

-1,171

-601

39%

-0.15

-0.11

16%

3,924

    Poor

214

28

52%

0.00

0.01

Source: HIECPS, Feb 2008 to Feb 2009 (CAPMAS 2009).

Men had larger mean and median declines in real earnings than women, with an average decline of 1,189 EGP versus 391 EGP, respectively. This is not entirely unexpected in a period of rising inflation as men were initially earning more. Column (6) and (7) show a cloudier picture when it comes to differences across gender in terms of the average change in log earnings, or the percentage change in earnings. Men had a slightly larger median decline, but a slightly lower average decline. Thus, the story is less clear when it comes to the difference in log earnings.

Column (1) of Table 8 provides a simple regression of earnings change on each variable alone. This allows an easy way to identify the difference between groups relative to the omitted category (denoted by a 0) and whether this differential is statistically significant. The 798 EGP difference between average earnings for men and women is statistically significant.
Table 8

Earnings changes by individual characteristics, with relation to initial earnings

 

Dependent variable

Change in earnings

Change in log earnings

  

(1)

(2)

(3)

(4)

 

Individual Characteristics

no controls

with income as a control

no controls

with log income as a control

Gender

    
 

Male

−798***

2,751***

0.06**

0.38***

Age

     
 

35-44

−344

1,218***

−0.01

0.06**

 

45-58

−1,059***

1,018***

−0.09***

0.01

Education

    
 

Basic education/literacy

−754*

1,181***

0.02

0.17***

 

Completed Primary

−1,290

1,155***

0.04

0.20***

 

Secondary education

171

1,709***

0.11***

0.25***

 

University degree

−70

3,513***

0.14***

0.37***

Urban

    
 

Urban

−789*

1,624***

0.03

0.16***

Region

    
 

Lower Urban

1,905**

1,096

0.10*

0.09*

 

Lower Rural

1,784**

−1,355***

0.02

−0.12***

 

Upper Urban

1,265

−560

−0.01

−0.07

 

Upper Rural

1,491*

−1,651***

−0.05

−0.20***

Type of worker

    
 

wage worker private

5

−1,263***

−0.07**

−0.14***

 

employer

−2,175***

−516

−0.27***

−0.19***

 

self employed

−444

−2,495***

−0.22***

−0.42***

 

unpaid worker

848**

−3,657***

0.02

−0.44***

Informal Worker

    
 

Formal

−236

2,279***

0.08***

0.27***

Industry

    
 

Manu/Const

−371

1,967***

0.09**

0.31***

 

Trade/Services

−102

2,335***

0.13***

0.35***

Poor household in 2008

    
 

Poor

1,385***

−1,053***

0.15***

0.01

Initial earnings quintile

    
 

2

−65

1,778***

−0.13***

0.29***

 

3

−448***

3,106***

−0.17***

0.45***

 

4

−1,481***

4,162***

−0.29***

0.49***

 

5 = highest

−5,593***

6,988***

−0.46***

0.59***

Predicted initial earnings quintile

   
 

2

−311**

1,306***

−0.07**

0.18***

 

3

−437**

2,887***

−0.03

0.40***

 

4

−887***

3,561***

−0.08**

0.42***

 

5 = highest

−3,219***

5,511***

−0.14***

0.52***

Notes: legend: * p < 0.10; ** p < 0.05; *** p < 0.01. Standard errors account for strata and clusters in sample survey design. Omitted categories are female, 25-34 year olds, illiterates/no schooling, rural area, metropolitan area, wage worker public, informal workers, agricultural workers, non-poor households, and lowest 20% quintile of initial earnings or predicted earnings.

Source: HIECPS, Feb 2008 to Feb 2009 (CAPMAS 2009).

Before reading too much into these average changes, one should consider controlling for initial earnings. Column (2) of Table 8 presents the regression results with initial earnings included as the sole additional right hand side variable. Compared to a female with the same initial earnings, a male is expected to have an earnings change that is 2,751 EGP better than his female counterpart. Thus, whether one considers this period a particularly bad period for male or female earners depends on whether we choose to consider male and female earners who are alike in their initial labor market outcomes or simply compare typical outcomes for each gender irrespective of their initial starting point. Similar analysis of the change in log earnings is presented in column (3), which has the average difference from the omitted group, and column (4) which includes log initial earnings on the right hand side. The minimal advantage that men had relative to women in the average change in log earnings expands dramatically (from 6 percent to 36 percent) once we control for initial log earnings.

Reviewing the results of Table 7 and columns (1) and (2) of Table 8, earnings declines were often larger for those groups who started with higher initial earnings, just as they were for men versus women. Examples include older workers, those in urban areas and the Metropolitan region, those who were employers in 2008, those who weren’t unpaid workers in 2008, and those who lived in poor households in 2008. Consistent with this, but to a lesser degree, those who were formal workers and those who started outside of agriculture also had smaller mean and median changes. Working against this general trend, those with university education experienced smaller average losses than those with some primary schooling or complete primary schooling and those in the lower urban area seemed to withstand large losses relative to others despite high initial earnings. Yet, this relationship generally flips once one controls for initial earnings (or expands further in favor of the initially favored).

Columns (3) and (4) of Table 8 re-affirm this relationship with log earnings. The main difference is that the initial comparison between the favored and less favored groups is not as consistent. Besides the mixed results by gender, there are some notable exceptions where the log earnings changes of those who have low initial earnings are higher—such as for agricultural workers, informal workers, and those in rural regions. Results for age, education, employer relationship, and poor/non-poor household in 2008 generally conform to our previous analysis. Yet in almost all these cases, the initially favored group has much better average earnings changes once we control for initial log earnings is controlled.

With this in mind, we return to the flexible estimation approach of locally weighted regressions, mapping the relationship between earnings changes and initial earnings for particular groups of individuals.

Figure 5 shows that men consistently have better earnings change experiences than women who started at the same initial earnings18. The reason men had worse average and median earnings overall is because they started at higher initial earnings in 2008. This is clear through the fact that the area of support is pushed further to the right for men as compared to women. Thus, although men had larger declines on average, one might suspect that they actually had an advantage in responding to the surge in inflation.
Figure 5

Change in log earnings based on initial log earnings by gender, Egypt 2008–2009. Source: HIECPS (Feb 2008 to Feb 2009 (CAPMAS 2009)). Notes: Locally weighted regression was used to determine the predicted change in log earnings based on log real earnings in 2008. Analysis was done separately by gender and then plotted on the same graph.

Figure 6 assesses the change in log earnings based on the initial sector of work. Again, the negative relationship between log earnings change and initial log earnings is relatively consistent across all groups. In this case, the relationship in Table 7 is not overturned when controlling for initial earnings. Those who started out working in agriculture have consistently worse earnings changes than those working in other sectors.
Figure 6

Change in log earnings based on initial log earnings by industry in 2008, Egypt 2008–2009. Source: HIECPS (Feb 2008 to Feb 2009 (CAPMAS 2009)). Notes: Locally weighted regression was used to determine the predicted change in log earnings based on log real earnings in 2008. Analysis was done separately by industry and then plotted on the same graph.

Figure 7 suggests that there is no consistent differential for those who were in poor households in 2008 compared to those in non-poor households once one controls for initial earnings. There may be some gap at the lower and upper initial earnings, but those in poor households have a slight advantage in the middle range of shared initial earnings. Thus, the earnings mobility disadvantages that some individuals face do not appear rooted in initial year poverty status.
Figure 7

Change in log earnings based on initial log earnings by 2008 poverty status, Egypt 2008–2009. Source: HIECPS (Feb 2008 to Feb 2009 (CAPMAS 2009)). Notes: Locally weighted regression was used to determine the predicted change in log earnings based on log real earnings in 2008. Analysis was done separately by 2008 poverty status and then plotted on the same graph.

Column (7) in Table 7 shows the regression coefficient from bivariate regressions of earnings change on the category of interest, controlling for initial earnings. By looking across columns (6) and (7), we can see that there are multiple cases where our impression of whether a group (such as males) has had better average outcomes than another (women), is overturned when we compare the two groups conditional on the same initial earnings. This overturning effect generally holds for females, younger workers, those living in rural areas, or in regions with lower initial earnings, unpaid workers, informal workers, those working in agriculture and those living in poor households. While mean and median earnings changes for these groups were better than those of their counterparts, they performed worse on average than their counterparts who had similar initial earnings.

In short, it is very important to consider the impact of initial earnings when evaluating how certain types of individuals were able to succeed and advance during the period. This type of evaluation generally requires panel data.

5.5 The role of transitions across types of work

Next we examine the systematic earnings changes associated with moving to a different sector of employment or to a job with a different employer relationship. First, we assess how often these transitions take place.

Table 9a shows that just more than half (51 percent) of all panel workers are employed in trade and services in 2008. Ninety-two percent of these workers are employed there again in 2009, while five percent are in manufacturing or construction and three percent are in agriculture. Agriculture also retains the vast majority (88 percent) of its workers in 2009 while manufacturing and construction work appears a bit less stable, with the sector retaining just 78 percent of 2008 workers in 2009. Overall the flows result in a slight loss of employment in agriculture and manufacturing/construction and movements into trade and services, which now employ 53 percent of workers. Overall, 12 percent of panel members were in a different sector in 2009 than in 2008.
Table 9

Employment transitions by type of work or worker, Egypt 2008 to 2009

a: Sector of work

   
 

Sector in 2009

  

Sector in 2008

Agriculture

Ind/Const

Trade/Services

  

Agriculture (31%)

87.8%

3.8%

8.4%

  

Industry/Construction (19%)

5.0%

78.2%

16.7%

  

Trade/Services (51%)

2.7%

5.0%

92.3%

  

Total

29%

18%

53%

  

b: Employer Relationship

  
 

Employer Relationship in 2009

 

Employer Relationship in 2008

Public Wage

Private Wage

Employer

Self-employed

Unpaid

Public Wage (33%)

94.9%

3.6%

0.8%

0.5%

0.1%

Private Wage (25%)

15.4%

66.0%

8.3%

8.1%

2.1%

Employer (15%)

2.8%

11.8%

69.3%

12.5%

3.5%

Self-employed (16%)

2.5%

11.4%

16.1%

55.3%

14.6%

Unpaid (10%)

2.2%

5.8%

3.5%

13.9%

74.7%

Total

36%

22%

16%

15%

11%

c: Formality

    
 

Status in 2009

   

Status in 2008

Informal

Formal

   

Informal (53%)

88.2%

11.8%

   

Formal (47%)

14.8%

85.2%

   

Total

53%

47%

   

Source: HIECPS, Feb 2008 to Feb 2009 (CAPMAS 2009).

Table 9b demonstrates the strength of tie to public wage employment. Ninety-five percent of public wage workers in the panel remained in public wage employment in 2009 as compared to just 66 percent of those in private wage work, 69 percent of employers, 55 percent of the self-employed and 75 percent of the unpaid workers. Flows from private to public wage work were particularly plentiful, helping to explain the three percent increase in public wage employment and the corresponding three percent loss in private wage employment in 2009, even though flows into public employment came from all other types of employment.

The transient nature of self-employment is also apparent as just 55 percent of 2008 workers were self-employed in 2009. More than ten percent of these individuals ended up as private wage employees (11 percent), as employers (16 percent), and as unpaid workers (15 percent). While moving to employer status might be thought of as a successful outcome for a self-employed individual, other outcomes may be less certain. Movement to private wage employment may often occur by choice and may often be the result of business failure. Movement to unpaid worker status might result from more family members working in a successful household enterprise or it might be a result of failure of the business that the individual was running and falling back into another household enterprise. The earnings mobility outcomes may help us form opinions on these.

Tables 10a-b and 11 should be interpreted bearing in mind the mean (−950) and median (−446) earnings losses for the population of workers as a whole.
Table 10

Mean and median earnings change by employment transition, Egypt 2008 to 2009

a: Sector of work

   
 

Sector in 2009

  

Sector in 2008

 

Agriculture

Ind/Const

Trade/Services

  

Agriculture

Mean

−852

−97

−921

  
 

Median

−456

−115

−222

  

% positive

38%

48%

45%

  
 

n obs

1,002

38

92

  

Industry/Construction

Mean

−1,342

−1,411

−154

  
 

Median

−963

−641

−258

  

% positive

32%

37%

42%

  
 

n obs

37

507

104

  

Trade/Services

Mean

−3,878

−966

−842

  
 

Median

−1,148

−947

−349

  

% positive

39%

42%

44%

  
 

n obs

54

87

1,559

  

b: Employer Relationship

 

Employer Relationship in 2009

Relationship in 2008

 

Public Wage

Private Wage

Employer

Self-employed

Unpaid

Public Wage

Mean

−614

−116

   
 

Median

−109

210

   

% positive

47%

51%

   
 

n obs

1,053

39

   

Private Wage

Mean

−640

−846

984

−178

−1,815

 

Median

−188

−641

808

−235

−1,661

% positive

47%

37%

61%

46%

12%

 

n obs

124

566

78

71

20

Employer

Mean

 

−2,956

−2,481

−3,441

 
 

Median

 

−2,364

−1,482

−1,744

 

% positive

 

15%

32%

23%

 
 

n obs

 

61

390

68

 

Self-employed

Mean

 

−3,388

19

−953

−1,116

 

Median

 

−1,719

−205

−471

−603

% positive

 

22%

47%

38%

33%

 

n obs

 

65

93

331

95

Unpaid

Mean

 

2,047

 

305

−277

 

Median

 

1,644

 

245

−230

% positive

 

93%

 

59%

39%

 

n obs

 

20

 

47

274

Notes: Cells with less than 20 observations not shown.

Source: HIECPS, Feb 2008 to Feb 2009 (CAPMAS 2009).

Table 11

Formality

 

Status in 2009

Status in 2008

 

Informal

Formal

Informal

Mean

−865

−640

 

Median

−479

−222

% positive

 

39%

46%

 

n obs

1,690

212

Formal

Mean

−2,848

−765

 

Median

−1,160

−267

% positive

 

29%

45%

 

n obs

239

1,340

Source: HIECPS, Feb 2008 to Feb 2009 (CAPMAS 2009).

Those who left agricultural work were generally better able to stem the earnings losses. The median change and the percent experiencing positive earnings gains were higher for those who switched sectors, though the average decline in earnings for those who moved to trade/services was actually larger than those who stayed in agriculture. Meanwhile, those who moved into agricultural work were experiencing larger than average earnings losses. This evidence is consistent with what one might expect if there is significant labor market segmentation, with agriculture representing the least preferred free entry sector and the differential pay being offered to the same individual working in different sectors. However, a much more detailed analysis would be needed to assess this hypothesis more fully.

The earnings decline for those remaining in the industry/construction sector was large. Combined with the previous information of workers flowing out of this sector, a consistent picture emerges that this sector was under stress during the 2008 to 2009 period. The only group of individuals that started in industry/construction and ended with less than average earnings losses was the group that moved to trade/services. Correspondingly, those who started in trade/services suffered high median losses if they moved to industry/construction, but the mean earnings loss for this group and the percent experiencing positive gains were basically the same as the overall average. Thus, the data is less clearly consistent with what we would expect if there was segmentation across industry/construction and trade/services sectors. It does appear that the industry/construction sector was under strain. It is a bit surprising that those who transitioned from industry/construction to trade/services had such relatively positive earnings changes. This seems to be driven by those coming from construction or electrical utilities rather than manufacturing sectors, but the estimates are too variable to find the differences to be statistically significant.

Public wage workers who stayed on in public wage employment did better than average in stemming the loss in real earnings. This is surprising given their high initial earnings levels and our earlier results regarding high initial earnings. Those who moved from public to private wage employment did even better. This suggests that such moves were likely voluntary moves.

In fact, with the exception of movements into unpaid worker status, wage workers who moved to new work categories consistently had better earnings changes than those who stayed within their same worker category. Thus, a generalization might be to assume that most observed changes of wage workers were voluntary, while those moving to unpaid worker status were involuntary. It does not appear that this was a period of substantial labor shedding of wage workers, public or private. Instead, workers absorbed losses in their real wage.

In contrast to this, the employed and self-employed who moved to private wage experienced even worse mean and median changes than those who stayed in their respective categories, consistent with a notion of being involuntarily pushed out of business rather than finding a better job.

Those who started as employers took large losses whether they moved sectors or tried to ride out the storm remaining as an employer. The self-employed also had relatively large losses. This might seem to be surprising given that such individuals have the opportunity to immediately change their product prices in response to inflation (and have their earnings rise) whereas workers may be locked into longer term contracts. This will be discussed more fully in Section 6.

Movements from formal to informal employment tended to result in very large losses. They were also quite numerous. This could signal labor shedding, where individuals are being forced to change jobs and losing out on pay. However, there was an inflow from informal to formal employment that was similar in size as the above outflow. This inflow did not result in earnings gains on par with the losses noted above. Therefore, it was not simply a case of individuals changing positions with similar earnings. An alternative explanation could be that employers under duress are opting out of the social pension system (and simultaneously giving workers reduced earnings). One might expect a certain flow from both of these causes in any time period. Without a baseline to know the typical flows and earnings changes from such transitions in Egypt, it is difficult to pinpoint an explanation.

Interestingly, those who moved from informal to formal sectors had earnings changes that were significantly but not so dramatically better than those who stayed informal. The changes were similar to those experienced by those who started and remained in the formal sector.

5.6 Multivariate analysis

We turn to multivariate analysis to better understand the determinants of earnings mobility. Table 12 presents a series of regressions that have the change in log earnings as the dependent variable. Results are presented using only exogenous base year characteristics and initial earnings (column 1) and then adding endogenous variables from the base year (column 2)19. Columns (3) and (4) include employment transition variables, with column (4) being the preferred specification (discussed below). In column (5), we replace initial log earnings with predicted initial log earnings and include all previous variables. In this specification, the effect of predicted initial log earnings is identified by fluctuations in log per capita consumption in the household in the base period.
Table 12

Determinants of change in log earnings, Egypt 2008 to 2009

 

(1)

(2)

(3)

(4)

(5)

Log earnings in 2008

−0.4655***

−0.5226***

−0.5352***

−0.5359***

 

Predicted log earnings in 2008

    

−0.2818***

Male

0.4423***

0.3258***

0.2734***

0.2828***

0.1474***

Age

0.0510***

0.0423***

0.0412***

0.0408***

0.0235**

Age-squared

−0.0006***

−0.0005***

−0.0005***

−0.0005***

−0.0003**

Married

0.1082**

0.1461***

0.1441***

0.1475***

0.0798**

Illiterate/no formal schooling (omitted)

     

Basic education/literacy

0.1130***

0.0417

0.0204

0.0214

0.0058

Completed Primary

0.1471***

0.0761**

0.0653*

0.0693*

0.039

Secondary education

0.2298***

0.1056***

0.0860***

0.0877***

0.0677**

University degree

0.4033***

0.2518***

0.2189***

0.2243***

0.1486***

Metropolitan (omitted)

     

Lower Urban

0.0723

0.0673

0.0627

0.064

0.0882

Lower Rural

−0.0764*

−0.0501

−0.0515

−0.0613

−0.0146

Upper Urban

−0.0754

−0.0722

−0.0861

−0.0898

−0.0623

Upper Rural

−0.1771***

−0.1417***

−0.1183**

−0.1327**

−0.0999**

Working Housewife

 

−0.2626***

−0.1251*

−0.1387**

−0.0996

wage worker public (omitted)

     

wage worker private

 

−0.0574**

−0.013

−0.0029

−0.0384

employer

 

0.0381

−0.0922

−0.0602

−0.1909***

self employed

 

−0.1214***

−0.0781

−0.0558

−0.0976

unpaid worker

 

−0.0447

0.2694***

0.2966***

0.3550***

Formal worker

 

0.0852***

   

Agriculture (omitted)

     

Manu/Const

 

0.0877**

 

0.0944***

0.060*

Trade/Services

 

0.0943**

 

0.0934**

0.068**

Poor in 2008

 

−0.0051

0.0053

0.0036

0.1438***

wage wkr public in 09 (omitted)

     

wage wkr private in 09

  

−0.0225

−0.0336

−0.0482

employer in 09

  

0.3171***

0.2557***

0.1958***

self employed in 09

  

−0.0619

−0.092

−0.0675

unpaid worker in 09

  

−0.4984***

−0.5534***

−0.5460***

Informal-informal (omitted)

     

Informal-formal

  

0.1049**

0.1210***

0.1135***

Formal-informal

  

−0.007

−0.0027

−0.0762*

Formal-formal

  

0.1470***

0.1559***

0.0517*

Agri - Agri (omitted)

     

Agri - Manu/Const

  

0.2356***

  

Agri - Trade/Services

  

0.1099

  

Manu/Const - Agri

  

−0.1032

  

Manu/Const - Manu/Const

  

0.1575***

  

Manu/Const - Trade/Services

  

0.1642**

  

Trade/Services - Agri

  

−0.1306

  

Trade/Services - Manu/Const

  

0.1405**

  

Trade/Services - Trade/Services

  

0.1548***

  

Constant

2.3032***

3.0926***

3.1953***

3.2530***

1.6737***

n

3481

3481

3480

3481

3481

R-squared

0.2358

0.2728

0.3471

0.3411

0.1297

Notes: legend: * p < 0.10; ** p < 0.05; *** p < 0.01. Standard errors account for strata and clusters in sample survey design. Column (5) uses bootstrapped standard errors.

Source: HIECPS, Feb 2008 to Feb 2009 (CAPMAS 2009).

Initial earnings are a powerful predictor of earnings change across all specifications. The coefficients imply that the predicted change in log earnings decreases by between 0.47 and 0.54 for each additional unit in reported log income. When using predicted initial log income, the coefficient drops to −0.28, which is still highly significant from an economic perspective. The coefficient is statistically significant at the one percent level for all specifications.

Another characteristic exhibiting a consistently important influence on predictions is gender. Males experience approximately 27 percent to 44 percent higher earnings changes than females, ceteris paribus, in columns (1)-(4)20. When using predicted initial earnings, this falls to about 15 percent higher, ceteris paribus, but remains both statistically significant and important in terms of economic loss. In contrast, re-running column (4) without including initial earnings, leads to a negative though statistically insignificant gender effect (see Table 13).
Table 13

Determinants of change in log earnings, with and without initial earnings

 

(1)

(2)

 

with initial earnings

without initial earnings

Log earnings in 2008

−0.5359***

 

Male

0.2828***

−0.0076

Age

0.0408***

0.0022

Age-squared

−0.0005***

−0.0001

Married

0.1475***

0.0001

Illiterate/no formal schooling (omitted)

  

Basic education/literacy

0.0214

−0.0308

Completed Primary

0.0693*

−0.0126

Secondary education

0.0877***

0.0203

University degree

0.2243***

0.0326

Metropolitan (omitted)

  

Lower Urban

0.064

0.1136*

Lower Rural

−0.0613

0.0461

Upper Urban

−0.0898

−0.0238

Upper Rural

−0.1327**

−0.0368

Working Housewife

−0.1387**

−0.0457

wage worker public (omitted)

  

wage worker private

−0.0029

−0.033

employer

−0.0602

−0.3161***

self employed

−0.0558

−0.0734

unpaid worker

0.2966***

0.4930***

wage wkr public in 09 (omitted)

  

wage wkr private in 09

−0.0336

−0.0455

employer in 09

0.2557***

0.1889***

self employed in 09

−0.092

−0.0678

unpaid worker in 09

−0.5534***

−0.5493***

Informal-informal (omitted)

  

Informal-formal

0.1210***

0.1174**

Formal-informal

−0.0027

−0.0812**

Formal-formal

0.1559***

0.0414

Agriculture (omitted)

  

Manu/Const

0.0944***

0.0117

Trade/Services

0.0934**

0.0277

Poor in 2008

0.0036

0.2309***

Constant

3.2530***

−0.0816

n

3481

3481

R-squared

0.3411

0.1228

Notes: legend: * p < 0.10; ** p < 0.05; *** p < 0.01. Standard errors account for strata and clusters in sample survey design.

Source: HIECPS, Feb 2008 to Feb 2009 (CAPMAS 2009).

Higher education levels also seem to be associated with higher earnings changes when controlling for initial earnings. Focusing on column (4), those with secondary education and university education had statistically significant differences in earnings changes compared to those who were illiterate/had no schooling, with gains of approximately 8.8, and 22.4 percent respectively. If one does not control for initial earnings, the estimated differences by education level are neither individually nor jointly statistically significant.

Age is also statistically significant, with earnings changes increasing and then decreasing with age. In column (4) the turning point is 43 years of age, while in column (5) it is 39 years of age (not shown in the table). Being married raises predicted earnings change by approximately 15 percent in the preferred specification, although, in column (5), the point estimate is less than half the size and it is only statistically significant at the ten percent significance level. These effects were also considerably different when running regression analysis without initial earnings21.

Using the preferred specification, the upper rural region had statistically lower earnings changes, ceteris paribus, than the Metropolitan region. Returning to Table 7, one may notice that the highest median pay in 2008 was in the Lower Urban region, followed by the Metropolitan region and the lowest pay was in Upper Rural region.

Looking at the employment transition variables, there are some surprising results. First, after controlling for other characteristics and initial earnings, the variables identifying transitions across industry in column (3) are not jointly statistically significant. The data did not reject the null hypothesis that only the initial sector of employment was needed for predicting earnings changes. This is surprising, but does allow a simpler form to be used as our preferred specification in column (4).

Using just initial industry characteristics we determine that those who started in agriculture experienced approximately nine percent lower earnings changes than others. The results are statistically significant. This result might seem surprising since the inflation was driven in part by rising food prices. One might assume that food producers would be able to increase their prices at least at the rate of inflation.

There are some potential explanations behind this unexpected result. We begin with some explanations that justify why farmers’ gains from rising food prices might be less than expected for those unfamiliar with the Egyptian setting. First, for most of the small farmers, the agricultural production makes up only up to 40 percent of their incomes, and they simultaneously sell their labor to other farmers/enterprises. Since higher food prices only benefit farmers on the share they sell, and agricultural wages may be slow to adjust, particularly if the food price shocks come too late to adjust the production size, the overall gains may be less than expected. Second, many farmers in Egypt are net consumers of food. Egypt is a very large importer of grains (mostly wheat), oil, and sugar—three main items in the consumption basket. Meat production relies heavily on imported grains as well. Therefore, it's possible that the prices for the main consumption items and input costs were going up substantially faster than the production items. Finally, even assuming that farmers could benefit from higher prices, it did not seem the case for most of the domestic prices on food items, which increase in a very different way from international prices, as shown in Figure 8.
Figure 8

Prices of imported and domestic food (percentage change January 2006-January 2008). Reprinted from World Bank, DECPG (2008)

These explanations are admittedly lacking. While they offer ideas on why the relative gains for those engaged in agriculture may be weaker than expected, they do not explain why those in agriculture had outcomes that were actually worse than others.

An alternative explanation suggests that farmers faced rising input costs and suppressed gains in output prices. The rising input costs are primarily attributed to steeply rising fertilizer costs. On the other hand, a ban on rice exports caused the domestic price of rice to be lower than expected. These storylines are presented here as one possible explanation that warrants much further examination. We do not provide supplemental evidence other than to say this proposed explanation is somewhat consistent with the fact that the large earnings declines for those initially in agriculture are predominantly driven by changes for those initial employers or self-employed rather than for those who are working in agriculture as wage workers. Future work will allow us to offer more concrete evidence to support or dispute this storyline22.

Those who ended up in the formal sector in 2009 (whether or not they were in the formal sector in 2008) had statistically significant changes in earnings, 12 to 16 percent greater than those who started informal and remained informal. Those who started in the formal sector and moved to informal sector did no worse than those who were informal throughout.

For employer relationship transitions, we did not include all possibilities as there were many cells with very few observations. Instead, we include category dummies in the base and final year. The data suggests that, ceteris paribus, there was no statistically significant advantage to where one came from, but that those who were employers in 2009 experienced significant higher earnings changes and those who were unpaid in 2009 had significantly lower earnings changes.

Overall, individuals from initially advantaged groups tended to suffer smaller losses in log earnings than their counterparts after we control for initial earnings. This result holds for such critical variables as gender, education, and region as well as for the married dummy variable and one’s industry of employment in 2008. If we do not control for initial earnings, these relationships are often reversed or statistically insignificant. With regard to employment transitions, there is evidence that those ending in the better employer relationship positions or in the formal sector in 2009, tended to end up with a higher predicted change in earnings. An individual’s transition across industries offers surprisingly little ability to help estimate his/her earnings change.

5.7 A review of changes by initial poverty status

Individuals from initially poor households fared as well or better than others. They do not appear to be more vulnerable to earnings declines during this period. Those initially from poor households had substantially better average and median earnings changes than those from non-poor households. While this relationship was reversed when conditioning on initial earnings alone, non-parametric analysis showed limited differences for individuals from poor rather than non-poor households (see Figure 7). More importantly, the initially poor variable showed no statistical significance in the multivariate analysis (see Table 12), except for the positive coefficient in the specification using predicted earnings (column 5). This was significant at the one percent level and suggested a 14 percent advantage in earnings changes for those from poor households, ceteris paribus.

5.8 A more detailed examination by gender

Table 14 offers separate regression analysis by gender. Some interesting differences emerge. First, while the importance of initial earnings is apparent for all three groups, the impact is greater for women as compared to men. This could be due to a variety of reasons. For example, women may have a higher degree of transitory income as a general norm. Or, it could result from period-specific changes. Suppose, for example, successful women had a disproportionate decline in earnings because households had to spend more on food and had less left to spend on stock for the female run enterprises. Simultaneously, women with low earnings may have had disproportionate gains in earnings as they worked more aggressively in the labor market in an attempt to make up for inflation reducing the buying power associated with the male breadwinner’s take home pay. We do not examine these points fully here, but leave them as open questions.
Table 14

Determinants of change in log earnings, by gender

 

Male

Female

Log earnings in 2008

−0.4879***

−0.6533***

Age

0.0354***

0.0351

Age-squared

−0.0004***

−0.0004

Married

0.2429***

−0.1033*

Illiterate/no formal schooling (omitted)

  

Basic education/literacy

0.0192

0.0206

Completed Primary

0.0515

0.1422

Secondary education

0.0747**

0.1005

University degree

0.2301***

0.2564***

Metropolitan (omitted)

  

Lower Urban

0.0744

0.0793

Lower Rural

−0.0448

−0.0411

Upper Urban

−0.0384

−0.1535

Upper Rural

−0.1243**

−0.1631*

wage worker public (omitted)

  

wage worker private

−0.0403

−0.1930*

employer

0.0024

−0.0382

self employed

−0.0206

−0.3898**

unpaid worker

−0.0503

−0.2502

Formal worker

0.0703**

0.2116**

Agriculture (omitted)

  

Manu/Const

0.0337

0.1795

Trade/Services

0.0260

0.2113**

Poor in 2008

−0.0160

−0.0198

Constant

3.2490***

4.4566***

n

2439

1042

R-squared

0.2672

0.3311

Notes: legend: * p < 0.10; ** p < 0.05; *** p < 0.01. Standard errors account for strata and clusters in sample survey design.

Source: HIECPS, Feb 2008 to Feb 2009 (CAPMAS 2009).

Additional differences exist between males and females. For example, while married men had 24 percent higher earnings gains than an otherwise similar non-married men, married women had 10 percent lower earnings than an otherwise similar non-married woman. However, the effect for females was not statistically significant. Additionally, females saw much greater differentials in earnings change experiences based on formal work status in 2008; employment type in 2008, specifically in comparison to public wage workers; and sector of employment in 2008, specifically non-agricultural versus agricultural workers.

While these differences are present, other aspects of the earnings change experience are quite similar across gender. Both groups were estimated to have earnings changes that increase and then decrease with age, with nearly identical turning points (at 43 years). Additionally, there were nearly identical gains for those with a University degree compared to those with no education. Other results across education levels were quite similar for males and females or were statistically insignificant. Results were also similar across gender when looking at changes experienced in different regions of Egypt.

6. Conclusions and areas of further research

Using panel data from February 2008 to February 2009, we observe changes in earnings experienced by workers in Egypt during a period of high inflation. In aggregate, the data shows that the labor market reacted to this rapid inflation by inducing a rapid decline in real earnings. There was no major reduction in employment levels, at least not for prime-aged workers. Average nominal earnings remained relatively unchanged. The mean and median changes in real earnings were almost universally negative across a wide spectrum of individual and regional characteristics.

At the individual level, however, the earnings change experience exhibited substantial heterogeneity, with 27 percent of the workers with employment in both periods reporting pay at least fifty percent higher or fifty percent lower than their reported 2008 earnings. Additionally, many individuals transitioned to new sectors of work or employer relationships during this period. Thus, for many individuals, this was a period of considerable fluidity and complexity compared to the rather simple aggregate story. As this is the first panel dataset in Egypt that captures such movements, we cannot determine the extent to which these earnings changes and employment transitions differ from what is typically found in a more stable economic period.

However, using this panel data and a descriptive regression approach, we are able to identify key covariates that have a significant association with earnings change. The descriptive regression approach cannot give us causal interpretations, but it does provide statistical determinants of earnings change. It gives circumstantial evidence that favors particular items as driving forces. It also leaves us with many new questions to explore in the future.

We found that initial earnings had a powerful negative association with earnings change. Those with higher earnings in 2008 had a greater average (and median) decline in earnings. This might be expected in a time period where inflation is driving the change in real earnings. Yet, the data also shows that those with higher reported log earnings in 2008 had a greater average decline in reported log earnings as well.

Such results may be driven by a spurious correlation resulting from classical measurement error. Thus, we re-computed the analysis using predicted earnings in place of initial earnings. The levels result was robust to this approach. The log result was not robust, although the coefficient was still negative. Thus, the data suggests that those with lower initial earnings experienced percentage declines that were, at worst, similar to those with higher initial earnings and there is some evidence that the percentage declines were lower for those with lower initial earnings.

The importance of controlling for initial earnings was evident when examining the relationship between earnings change and gender. On average, men experienced a much larger average decline in earnings than women. However, for men and women with the same initial earnings, women had a greater average fall in earnings.

For a number of other covariates, the story was similar. The disadvantaged groups—such as young workers, those living in rural areas or those living in low-earning regions—experienced smaller average declines in earnings than their advantaged counterparts. However, if individuals started with the same initial earnings, it was members of the disadvantaged group that experienced greater average declines. One exception to this was that those from poor households did not generally show themselves to be more vulnerable to earnings declines even after conditioning on initial earnings.

When controlling for a larger host of covariates, higher initial earnings remained negatively related to earnings change. This association is statistically significant and has large economic significance as well whether controlling for reported or predicted initial log earnings.

The inevitable question is, why? What is it about having high initial earnings in 2008 that causes individuals to fall further relative to similar individuals with lower initial earnings? Did these individuals possess some unobservable characteristics that are no longer being rewarded as well as before? Is it capturing a dynamic adjustment process where individuals with positive or negative shocks gradually return to their conditional mean? To begin to address these questions, we would need panel data with more than two periods. Thus, it remains an open question for future work.

One consequence of this negative relationship between earnings change and initial earnings is that individuals are often changing position in the earnings distribution. Thus, when assessing income inequality over time using cross-sectional data, one should recognize that there are likely different people in different earnings positions over time (see Fields and Sanchez-Puerta (2010) for a clear graphical example).

There is also more work to be done in understanding why women had worse earnings change experiences than men who started at the same initial earnings level. This result was robust to excluding women who were listed as working housewives. Yet, it is worth investigating whether these changes were brought on by lower investment in female businesses, firms offering lower pay or holding pay fixed for women or other possible reasons. Additionally, understanding why earnings changes were more responsive to initial earnings for women than men may help us better understand how initial earnings impact earnings change.

This influence of initial earnings leads to a situation where various groups might see themselves as disproportionately affected by the inflation shock and ensuing decline in earnings. For example, men might recognize the absolute value and even percentage value of their losses were greater than that of females. On the other hand, women might notice that their losses and percentage losses were far greater than men who started with the same initial earnings (and had otherwise similar characteristics as them). It is an open question whether such settings give rise to more social discontent than settings where aggregate differences and differences conditioning on those like oneself allows only one group to claim relative disadvantage. The absolute decline in earnings experienced by so many different types of individuals was undoubtedly a source of widespread social discontent in this period. But a hypothesis worth considering is that such discontent becomes even greater in situations where all groups could also consider themselves relatively worse off depending on whether they focus on aggregate effects or effects conditional on personal characteristics.

Finally, earnings changes for those initially engaged in agriculture were unexpected, given that the changes in inflation were driven in part by rising food prices. Those who started in agriculture had the biggest average decline in earnings, holding all else constant. A number of possible explanations have been raised. In our view the most plausible revolve around rising fertilizer costs with relatively stagnant farm-gate prices for at least some crops. A more complete explanation would require a detailed examination of the agricultural sector and is left for future work.

Endnotes

1Concurrent analysis of earnings mobility using this data has been undertaken by Silva as part of the World Bank investigation into informality. Earlier, Assad and Roushdy (2007) also used the panel data from the Egyptian labor market conditions surveys (ELMPS) conducted in 1998 and 2006 together with the 2005 HIECS to provide very detailed information on employment status, earnings, and poverty. The work was descriptive using transition matrices and focused on wage workers. Their results reveal that between 1998 and 2006 there was a notable improvement in labor-market conditions in Egypt. New jobs were created at a faster rate than the growth of the labor force, leading to a reduction in the unemployment rate. They also found an uneven increase in earnings where only few of the poor had benefited from these gains. However, both studies did not take into account the relationship between the initial earnings and earnings changes that is central to this paper.

2This statement is not meant to be a central thesis for what kept inflation from continuing to increase. The lack of adjustment of wages, itself, was likely driven by many underlying factors.

3An alternative hypothesis would be that they did raise prices but that demand for their product fell as buyers had less of their constant nominal earnings to spend on domestic services and products due to increased prices of the imported goods that they were buying.

4This process divides enterprise gains solely among residents engaged in a home enterprise. Such a division can lead to measurement error. Fortunately, more than half of those engaged in household enterprise work are the only ones in the household engaged in such activity (and thus do not need to split profit). In multiple person enterprises (often two person activities), there may be some concern in this regard.

5We also discard 9 observations in 2008 and 12 observations in 2009 that had zero earnings despite being listed as employed. These were primarily unpaid workers and we believe the values to be erroneous.

6We can reject the hypothesis that each dummy variable equals zero when testing them separately. Surprisingly, using a ten percent significance level, we cannot reject the joint hypothesis test that both of these dummy variables simultaneously equal zero.

7Such descriptive regressions do not attempt to eliminate or fix endogenous right hand side variables that can conflate causal interpretations of regression coefficients. This approach is useful in predicting who was affected by a given change, without explaining why. In other words, we may be able to say that individuals who transitioned from sector A to sector B experienced higher average earnings changes than those of similar age, education level, etc. who stayed in sector A. We cannot say that they earn more because they switched sectors or that anyone else who had made a similar switch would have enjoyed the same gain.

8Here, we ignore the fact that there is not a direct correspondence between earnings and consumption per capita.

9There may be some concern about marital status being considered exogenous. This will be discussed in the results section.

10The construction of the poverty lines is explained in The World Bank (2011).

11Cameron and Trivedi (2005) also offer a useful discussion of the lowess estimator. However, the previous paragraph paraphrases from the STATA manual, which also includes the basic formula for the tri-cube weighting scheme: wj = and Δ = 1.0001*max(xi+ - xi, xi – xi-) where xi+ and xi- are edges of the usable observations. As our bandwidth is 0.8, this is limited to be no more than 40% of the observations above and no more than 40% of the observations below the current observation when observations are lined up according to value of x. Put another way, an observation at the median level of x would use the middle 80% of the sample in its regression, while an observation at the upper extreme of x would use just the observations with the highest 40% of values of x. In our case x is initial earnings level. One negative feature of this approach is that it does not incorporate the probability weights in the underlying sample.

12In the case where we start with a negative coefficient, the attenuation bias will help mitigate the overall bias. For example, if the actual value is −0.50 rather than 0.50, we would find −0.55.

13Note: The measurement error, μ08, will no longer be present in the predicted earnings from 2008, thus eliminating the spurious correlation. The new equation will be: Δyi + (μ09,i – μ08,i) = β0 + β1(yhat08,i) + ϵi. As the long as the measurement error in 2008, μ08,i is not correlated with any of the variables used to predict initial earnings, which is a standard assumption of classical measurement error, it will not be present in our predicted earnings for 2008, here denoted by yhat08,i.

14Employment participation rates were also increasing from 48% to 53% among those 18 to 24 year olds in 2008 and the rate for those 18 to 24 years old in 2009 was also higher at 49%. These conclusions assume that those who lost their jobs were not more likely to be missing from the sample in 2009.

15Our data does not allow us to know if the individual is in the same job in both periods. We will also review changes in sectors and relationship with employer later in this paper.

16The difference in logs of (+ or -) 0.5 is a very rough approximation of a 50% change in earnings. Twenty seven percent of the weighted sample had reported earnings in 2009 that were 50% greater or less than reported 2008 earnings.

17In our data the variance of initial log earnings is roughly 0.68. Thus, if the variance of μ is .204 or greater, it would imply that the true coefficient is non-negative. Given a Normality assumption on our classical measurement error term and a standard deviation of 0.2040.5, this implies that 32 percent of respondents would be giving answers that were at least 45 percent above or below their true earnings level.

18See Figure 9 for the lowess model of the change in earnings after trimming the data of the lowest and top five percent of observations for each gender. Figure 10 shows the diagram using a local mean rather than local regression.
Figure 9

Change in earnings based on initial earnings by gender, Egypt 2008 to 2009. Source: HIECPS (Feb 2008 to Feb 2009 (CAPMAS 2009)). Notes: Locally weighted regression was used to determine the predicted change in earnings based on real earnings in 2008. Analysis was done separately by gender and then plotted on the same graph. The top and bottom 5 percent of initial earnings were excluded for each gender.

Figure 10

Change in earnings based on initial earnings, by gender, Egypt 2008 to 2009. Source: HIECPS (Feb 2008 to Feb 2009 (CAPMAS 2009)). Notes: Local mean constructed using tri-cube weight using lowess command with mean option. Top and bottom 5 percent of initial earnings are excluded for each gender.

19As mentioned earlier, we take exogenous to mean characteristics that these 25 to 58 year olds brought to the labor market in 2008. One might reasonably question whether the married variable is exogenous. We ran this specification without the married variable and found results on all other coefficients and corresponding standard errors to be extremely similar to the model presented in Table 12.

20A more accurate approximation would be between 35 and 63 percent (Ex. e0.30 - 1 = .35). However, we will follow convention and use the log approximation for all values in this paper.

21The age turning point was roughly 13 years old and being married had a zero coefficient.

22The authors would like to thank Ruslan Yemtsov, Julian Lampietti, and Maurice Saade for offering us some starting points in this area which we hope to follow in the future.

Annex 1: measurement error simulation

In order to demonstrate how our lowess simulations are affected by classical measurement error, we conducted a simulation. For simplicity we treat the observed values of log 2008 earnings as true values. In order to create our simulated values of log 2008 reported earnings, we add a simulated measurement error realization to each observation. These realizations are drawn randomly from a Normal distribution with a mean of zero and a standard deviation of 0.3. This standard deviation is admittedly arbitrary, but we feel it offers a significant amount of measurement error. Under this distribution, 32 percent of the sample would receive a random measurement error term with a magnitude greater than approximately 30 percent of the value of their ‘true’ earnings level. Five percent of the sample would receive measurement error that was more than approximately 60 percent of their ‘true’ earnings level.

We then constructed a second set of values in the exact same manner and called them log 2009 reported earnings. We take the difference and run a lowess regression of the simulated difference on the simulated 2008 data. This simulation thus represents what the lowess graph would look like if individuals had no real change in earnings but a fairly significant amount of measurement error. The result, shown in Figure 11 below, is initially disturbing as it shares many feature with our observed lowess results of the change in earnings on initial earnings. It is distinctly downward sloping.
Figure 11

Simulated change in log earnings based on initial log earnings, Egypt 2008 to 2009. Source: Authors’ compilation from HIECPS (Feb 2008 to Feb 2009 (CAPMAS 2009)). Notes: Simulated earnings values for 2008 and 2009 were both based on reported log 2008 earnings plus some random measurement error, which was drawn from a Normal distribution with mean zero and variance of 0.3. A change in earnings was then constructed directly from these simulated values. Finally, locally weighted regression was used to determine the predicted change in log earnings based on log real earnings in 2008. Upper and lower confidence intervals are shown as well.

However, this level of measurement error definitely cannot explain the results observed in our data. Figure 12 graphs the simulation again, this time also graphing the lowess graph using the change in log reported earnings and log initial reported earnings observed in our dataset. The simulation was downward sloping but the magnitude of the changes is tiny compared to the magnitude using our observed data. Thus, we are reassured that our lowess results are not driven solely by classical measurement error. Others may not feel as confident in this conclusion if they feel measurement error has a much larger variance or is otherwise different from our maintained assumptions. In this sense, this validation approach is similar to the approach of Fields et al. (Fields et al. 2003a).
Figure 12

Simulated and reported change in log earnings based on initial log earnings, Egypt 2008 to 2009. Source: Authors’ compilation from HIECPS (Feb 2008 to Feb 2009 (CAPMAS 2009)). Note: Simulated earnings values for 2008 and 2009 were both based on reported log 2008 earnings plus some random measurement error, which was drawn from a Normal distribution with mean zero and variance of 0.3. A change in earnings was then constructed directly from these simulated values. Finally, locally weighted regression was used to determine the predicted change in log earnings based on log real earnings in 2008. Upper and lower confidence intervals are shown as well.

Abbreviations

CAPMAS: 

Central Agency for Public Mobilization and Statistics

EGP: 

Egyptian Pound

HIECPS: 

Household Income, Expenditure and Consumption Panel Survey

HIECS: 

Household Income, Expenditure and Consumption Survey

LFSS: 

Labor Force Sample Survey

LPM: 

Linear Probability Model.

Declarations

Acknowledgements

The authors would like to thank Rania Rouchdy, Ruslan Yemtsov, Joanna Silva, Julian Lampietti, Maurice Saade, Heba El-Laithy, Sherine El-Shawarbi, the Editor and an anonymous referee for their comments. We would also like to thank Nicholas Diebel, Mallimalika Gupta, Choon-Sung Lim, and Caleb Pomeroy for research assistant work related to this paper. Ideas expressed in this paper are those of the authors and do not reflect policies of their respective organizations. The usual disclaimers apply.

Responsible editor: David Lam

Authors’ Affiliations

(1)
Economics Department, Boston College
(2)
Faculty of Economics and Political Science, Cairo University
(3)
World Bank

References

  1. Arabsheibani R: Male–female Earnings Differentials in an Islamic Country: The Case of Highly Educated Egyptians. Educ Econ 2000,8(2):129–138. 10.1080/096452900410703View ArticleGoogle Scholar
  2. Assaad R, Roushdy R: Labor market trends and living standards in Egypt. In Arab Republic of Egypt – Poverty Assessment Update, vol. 1. Arab Republic of Egypt: The World Bank; 2007:69–91.Google Scholar
  3. Beegle K, De Weerd J, Dercon S: Migration and Economic Mobility in Tanzania: Evidence from a Tracking Survey. Rev Econ Stat 2011,93(3):1010–1033. 10.1162/REST_a_00105View ArticleGoogle Scholar
  4. Belhaj-Hassine N: Inequality of opportunity in Egypt. World Bank Econ Rev 2012,26(2):265–295. 10.1093/wber/lhr046View ArticleGoogle Scholar
  5. Cameron AC, Trivedi PK: Microeconometrics: Methods and applications. Cambridge: Cambridge University Press; 2005.View ArticleGoogle Scholar
  6. CAPMAS (Central Agency for Public Mobilization and Statistics): Various issues. HIECS (Household Income, Expenditure and Consumption Surveys). Cairo, Egypt: CAPMAS; 2009.Google Scholar
  7. Cichello PL, Fields GS, Leibbrandt M: Earnings and employment dynamics for Africans in post-apartheid South Africa: A panel study of KwaZulu-Natal. J Afr Econ 2005,14(2):143–190. 10.1093/jae/eji001View ArticleGoogle Scholar
  8. Cleveland WS: Robust locally weighted regression and smoothing scatter plots. J Am Stat Assoc 1979, 74: 829–836. 10.1080/01621459.1979.10481038View ArticleGoogle Scholar
  9. Datt G, Olmsted J: Induced wage effects of changes in food prices in Egypt. J Dev Stud 2004,40(4):137–166. 10.1080/00220380410001673229View ArticleGoogle Scholar
  10. Fields GS: A brief review of the literature on earnings mobility in developing countries. 2008.Google Scholar
  11. Fields GS, Sanchez Puerta ML: Earnings mobility in times of growth and decline: Argentina from 1996 to 2003. World Dev 2010,38(6):870–880. 10.1016/j.worlddev.2010.02.001View ArticleGoogle Scholar
  12. Fields GS, Cichello PL, Freije S, Menéndez M, Newhouse D: For richer or for poorer? Evidence from Indonesia, South Africa, Spain, and Venezuela. J Econ Inequal 2003,1(1):67–99. 10.1023/A:1023990705770View ArticleGoogle Scholar
  13. Fields GS, Cichello PL, Freije S, Menéndez M, Newhouse D: Household income dynamics: A four country story. J Dev Stud 2003,40(2):30–54. 10.1080/00220380412331293757View ArticleGoogle Scholar
  14. Lundberg S: The added worker effect. J Labor Econ 1985,3(1):11–37. 10.1086/298069View ArticleGoogle Scholar
  15. Roushdy R, Gadallah M: Labor Market Adjustment to the World Financial Crisis: Evidence from Egypt. ERF Annual Conference paper, ERF17AC_054. 2011. http://www.erf.org.eg/cms.php?id=publication_details&publication_id=1363 Google Scholar
  16. Said M: The Fall and Rise of Earnings and Inequality in Egypt: New Evidence from the Egypt Labor Market Panel Survey 2006. Ch. 2. In The Egyptian Labor Market Revisited. Edited by: Assaad R. New York and Cairo: American University in Cairo Press; 2009:53–81.Google Scholar
  17. Said M: Wage Differentials during Economic Reform and Crisis: Evidence from Egypt and Jordan. Comp Econ Stud 2012, 54: 65–101. 10.1057/ces.2012.5View ArticleGoogle Scholar
  18. Serneels P: The added worker effect and intra household aspects of unemployment. CSAE Working Paper Series 2002–14. University of Oxford: Centre for the Study of African Economies; 2002.Google Scholar
  19. World Bank: Poverty in Egypt 2008–2009: Withstanding the global economic crisis. Egypt: The World Bank; 2011.Google Scholar
  20. World Bank, DECPG: Commodity Markets Review, Development Prospects Group. Washington, DC: The World Bank; 2008.Google Scholar

Copyright

© Cichello et al.; licensee Springer. 2013

This article is published under license to BioMed Central Ltd. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.