Gender wage gaps, informality and discriminatory social norms : empirical evidence from Uganda

(1)

1

Gender wage gaps, informality and discriminatory

social norms: empirical evidence from Uganda

Lizzie Dipple – MSc Economics Thesis

Development Economics track

Universiteit van Amsterdam

15th August 2016

Abstract:

In Uganda, as in many developing economies, informal employment is prevalent. Using Ugandan household survey data to evaluate a model of employment status selection that differentiates between formal and informal work, I uncover several differences in the way characteristics drive selection for men and women. In addition, discriminatory social norms appear to be associated with an increased likelihood of non-wage-earning status for women but also, unexpectedly, with an increased likelihood of formal work. Adjusting for the gendered selection process is key – the inclusion of selection corrections into a Mincerean wage equation increases the gap in formal wages significantly and alters the wage models, particularly for men. The gender gaps in selection-adjusted informal wages for both employees and the self-employed become insignificant.

Author: Elizabeth Dipple UvA student number: 11085185 Supervisor: Dr. R. Oostendorp Second reader: Dr. H. Oosterbeek Academic Year: 2015/2016 15 ECTS

(2)

2

I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

(3)

3

1. Introduction

The informal economy is a significant proportion of most developing economies, for example accounting for 66% of GDP in Sub-Saharan Africa, and in many regions informal employment is a greater source of employment for women than for men (Kolev & Sirven, 2010; Vanek et al, 2014). Women everywhere are at greater risk than men of being in low-paid work and informal employment tends to pay less than formal (Grimshaw, 2011). Hence for countries with a large informal economy there is a small but growing literature considering how informality and gender wage gaps are related. These wage gaps matter – aside from concerns over basic equality, there is also an efficiency argument since differences in productivity lead to differential pay for men and women, so potentially economically successful women may be stuck in a low productivity trap (World Bank, 2012).

Labour force participation and selection processes into employment types are known to be gendered (World Bank, 2012) and one potential contributor is the role of discrimination (Kantor, 2009). Many attempts to quantify discrimination rely on indicators of discrimination outcomes, such as gender differences in education, workforce participation, pay, political roles and executive power. However, the OECD’s recent collection of data on discriminatory social institutions measures underlying discrimination rather than simply its effects, by looking at societal norms, attitudes and practices (OECD, 2010). One in-country study has been completed, in Uganda, and hence that is where I base my research. Analysing how discrimination may fit into the model of employment choice is a new step towards understanding the deep drivers of gender gaps. Gender equality and female empowerment have become increased priorities in global development thinking and form the fifth of the UN’s recently unveiled Sustainable Development Goals.

Uganda had a female labour force participation rate of 75.7% in 2014, compared to a 79.2% participation rate for men, and the total Ugandan informal employment rate was 94.5% for women (as a percentage of total non-agricultural employment) and 92.7% for men.1 These are not large gaps but the gender wage gap will be more significant due to differences in educational achievement and given the high level of informality, for which wages are more volatile (Gunther & Launov, 2012). Gender wage gaps around the world tend to be smaller with greater economic development (figure 1) but, according to the data here, Uganda has a large gap even for its low per capita GDP, which I want to investigate.

Given that the informal economy involves the majority of workers in Uganda, and is sizeable across all of Sub-Saharan Africa as well as in many other low-income countries, quantifying wage gaps within informal employment is a useful step on the path towards improving economic growth and poverty alleviation. Within-household inequalities and differential gendered spending choices mean that increasing women’s earning possibilities is necessary for long-term poverty, health and social improvements (Saba Arbache, Filipiak & Kolev, 2010) as well as overall growth. Then, according to which model of informality is believed to hold, understanding any differences between formal and informal gender wage gaps could help with targeting equality-boosting interventions.

(4)

4

Figure 1: Gender wage gaps against GDP around the world

Sources: (i) ILOSTAT – unadjusted gender wage gap (the difference between average earnings of men and average

earnings of women expressed as a percentage of average earnings of men) based on household or labour force surveys in each country, for latest year available to 2015; (ii) World Bank World Development Indicators – GDP per capita, PPP (current international $) for latest year available to 2015; (iii) Uganda gender wage gap from author’s own calculations based on the 2011/12 Uganda National Panel household survey data.

In this thesis I quantify the gender wage gaps in formal and informal employment in Uganda and uncover the part explained by wage-structure discrimination. In order to account for selectivity, I model the process of selection into employment statuses. Thus I investigate whether there is gender segregation, with women and men encountering disparities in selection. I also consider how for women discriminatory social institutions fit into selection.

My results show that the employment status selection processes for men and women in Uganda are distinct. There are differing gendered effects according to personal and household characteristics on the probabilities of: working formally, working as informal employees, informal self-employment and being a non-wage-earner. Results suggest a positive association between social discrimination and the likelihood of women being non-wage-earners, but also of women being formally employed. Discrimination also seems to be positively related to the selection margins between formality and informality; hence there is no evidence that discrimination acts as a barrier to female formal employment.

I find that men’s and women’s wages in each employment type are also driven by differing factors and different scales of effects, with industry sector particularly significant for formal workers and the informal self-employed, and geography significant for both classes of informal worker. Adjusting for selection into employment status diminishes the effect of education on wages to insignificance for formal and informal employee men, but education remains significant for all women’s wages. The adjustment inflates what was a smaller gender gap in observed wages for formal workers so that the formal gap becomes the only significant gap in wage offers (potential wages). Correlations between unobserved factors driving selection and unobserved factors driving wages result in positive selection of

(5)

5

women into formal employment on average and negative selection of men into formal employment on average. My findings indicate no significant gender gap in potential wages in Uganda’s informal economy, neither for employees nor for the self-employed. Finally, considering the gendered effect of proportions across the employment types separately to the selection process accounts for about a quarter of the overall combined gender wage gap.

The remainder of the thesis is structured as follows. Section two gives the theoretical background to informality and gendered employment differences and considers the literature relating the two topics together. Section three outlines the methodology that I follow and section four describes the main data. Section five details the results from the various parts of the analysis, then section six concludes.

2. Background and theory

2.1. Informal employment

There are a number of definitions of informality, but the International Labour Organisation (ILO) now considers informal employment as the key concept – that is, both working in informal sector enterprises (whether as an employee or self-employed) and working informally for a formal sector enterprise. Informal jobs are those which, in law or in practice, are not subject to labour legislation or income taxation, and which do not provide social protection or employment benefits (ILO, 2003). Formality thus is a basic indicator of the quality of employment. Informal wages and earnings are usually lower than those of formal jobs (Grimshaw, 2011) and people working informally on average have less education and their returns to education are lower (Nordman, Rakotomanana & Robilliard, 2010; Gunther & Launov, 2012).

Chen (2012) lays out four main schools of thought regarding the informal economy. The dualist view (Hart, 1972) describes informality as due to the exclusion of some part of the population from modern economic opportunities in a formal sector. Under this segmented labour market, barriers to formal employment push the less well-endowed or more discriminated against into lower-paying informal work. However, there is evidence from Latin America that there may actually be an informal sector wage premium in Mexico and that informal returns to education are surprisingly high there (Marcouiller, Ruiz de Castilla & Woodruff, 1997).

Under the structuralist school (Castells & Portes, 1989), informality is based on capitalism seeking low-cost means of production and hence fomenting cheap informal labour. However, according to Becker (1971), increased competition should actually decrease discrimination, as it is costly to employers. Since women tend to earn less compared to men in general (Kolev & Sirven, 2010; Grimshaw, 2011), economic growth would increase women’s labour force participation but this would increase the gender wage gap, unless correcting for selection into remunerated work. However, Warnboye & Seguino (2015) find mixed effects of trade expansion on women’s employment in their cross-country study of Sub-Saharan Africa.

(6)

6

The legalist school of thought (De Soto, 1989) sees a hostile legal regime increasing the enterprise costs of formality. Gendered lack of access to finance could be a barrier to formality for self-employed women, although this does not necessarily say anything about gender wage gaps. Aterido, Beck & Iacovone (2013) relate that, despite a reported high degree of business discrimination against women in Sub-Saharan Africa and finding an unconditional gap in access to finance, this gap disappears once characteristics – including a selection effect for whether businesses are formal – are controlled for. This suggests that gender wage gaps within informal and formal employment would not be due to difficulties accessing finance.

The final view is the voluntarist school (Maloney, 2004), with a focus on the self-employed and micro-entrepreneurs, under which operating informally is a choice after weighing up costs and benefits. Selection into informal employment on individual preferences means that if, for instance, increased flexibility from informality for women (perhaps with household duties and/or children) is of greater importance than for men, then there will be an informal gender wage gap. Conversely, under this view, if the non-monetary benefits of formal employment, such as greater worker protection and maternity leave, are of greater importance to women than to men, then women would accept lower wages and this would predict a formal gender wage gap. Flexibility is often seen as one of the positives of informal jobs for women and, at least in developed economies, work in flexible jobs actually decreases the gender wage gap (Goldin, 2014).

Recent thinking focuses on two hypotheses: (i) the segmented labour market, where informal employment is a strategy of last resort to avoid unemployment when there is no social welfare safety net and formal employment is rationed in some way, and (ii) the competitive labour market, where choices are made because of desirable aspects of informality or simply an informal comparative advantage. Gunther & Launov (2012)’s findings in Côte d’Ivoire support neither outright, instead suggesting heterogeneity in the informal labour market and differing reasons – voluntary and involuntary – for entry. This mixed view is supported by Chen (2012) and Perry et al (2007). However, in his study on married women in Colombia, Magnac (1991) cannot reject the hypothesis of competitive, as opposed to segmented, labour markets.

2.2 Gendered differences in economic activity

According to the World Bank (2012), the main driver behind lower earnings and productivity for women compared to men is differences in economic activities. Chen (2012) describes a hierarchical model of informality whereby men are in the less-risky and better-paid types of informal employment. This predicts a gender wage gap within informal employment, but makes no prediction over its size relative to any corresponding formal wage gap. Gendered division of labour within households (Becker, 1985; Kolev & Sirven, 2010) means that women with greater family and care commitments work in less productive and lower-paid jobs, with comparative advantage between men and women intensifying any initial gender discrimination. Differences in productivity in different sectors can explain much of gender

(7)

7

wage gaps (Blau & Kahn, 1997), suggesting that any gaps will be different for informal and formal employment.

A multi-choice model for different employment statuses employed by Kantor (2009) in India concludes that social norms keep women in home-based work. Antecol (2000) provides evidence that cultural factors are strong drivers of the labour force participation gender gap and Fernández & Fogli (2006) argue that a woman’s work and fertility outcomes are influenced by her (cultural) heritage. These two papers are of particular interest here since they discuss the effect of cultural norms across many countries, not just developed economies, by using data on immigrants to the United States. Although supply and demand factors drive women’s labour force participation – such as increased supply due to rising female education and declining fertility, and increased demand due to export-oriented sector growth fuelled by trade openness (Chen et al, 2013) – social norms may weaken the connection between incentives for women to work and the actual outcomes observed (World Bank, 2012). Increased participation does not necessarily lead to a reduced wage gap. Chen (2012) argues that people can end up working informally due to wider structural forces and that social conditioning, informal ‘regulations’ and tradition are additional drivers. In their Chinese study, Chen et al (2013) conclude that there is gender segregation in job allocation within firms and that women are pushed into low value-added work. In Uganda, Campos et al (2015) find that women need psychosocial assistance rather than improvements in human capital to overcome or get around the norms that uphold gendered occupational segregation.

Following these arguments, I postulate that a social discriminatory effect may be one explanatory factor of gendered types of labour force participation in informal as compared to formal employment. According to OECD (2010), women’s participation in wage employment is lower in those countries with high levels of discrimination in the family context (parental authority, early marriage/polygamy and inheritance rights indicators). Lower access to land and credit is correlated with greater likelihood of women working as (informal) contributing family members. Societal norms about labour market status, such as unemployment, do seem to affect utility, and hence affect any selection into such statuses (Clark, 2003), although Clark finds the effect is stronger for men than for women.

2.3 Literature considering informality and gender together

There are many existing studies in the literature investigating gender wage inequality and others consider the differences between formal and informal sectors, but only a handful of quantitative studies link informality (or similar forms of poorly protected employment) and gender. Ben Yahmed (2013) looks at Brazil; Tansel (2000) at Turkey; Nordman, Rakotomanana & Robilliard (2010) at Madagascar; Gunther & Launov (2012) at Côte d’Ivoire; and Deininger, Jin & Nagarajan (2013) at India. All these studies consider the male– female wage gap separately for the different labour market segments and make some sort of adjustment for selection. Their results differ. Both Tansel and Ben Yahmed find a significant wage differential only for formal-type workers when controlling for selection into employment. However, Deininger, Jin & Nagarajan, Nordman, Rakotomanana & Robilliard and Gunther & Launov find exactly the opposite result, that is a larger wage differential or

(8)

8

significant only in informal-type employment. The various decompositions employed by Deininger, Jin & Nagarajan all conclude that the majority of the wage gap is down to discrimination rather than differences in observable characteristics (productivity gaps) or selectivity. The mixed results overall suggest that the country and/or context is significant.

Appropriately modelling selection into employment status is essential (Pradhan & van Soest, 1995); selectivity corrections in these types of analyses sometimes significantly change the gender wage gaps. Like Ben Yahmed (2013), I control for selection into different employment statuses using an updated Dubin–McFadden process (Bourguignon, Fournier & Gurgand, 2007) – although I only consider four statuses. I also extend the analysis by combining the data for men and women according to proportions working in the different employment types to uncover further aspects of the wage gap decomposition. The selection model includes sub-regional indicators to enable consideration of whether social norms and practices are correlated with gendered selection into the employment statuses.

This research adds to the existing literature in several ways. First, it applies recent advances in gender wage gap analysis techniques to formal and informal employment in a new country. Second, it combines the analysis of formal and informal employment for each gender to extend standard decomposition techniques beyond what has been done to date, as far as I am aware. And finally, it considers true measures of gender discrimination (as opposed to discrimination outcomes) in relation to a selection model for the first time.

3. Methodology

Following traditional human capital theory, the wage equation accounting for observable endowments follows the standard Mincerean form for an individual n in employment status j (j = formal, informal employee, informal self-employment or other), with a vector of explanatory characteristics: age (proxying for experience), age2, years of education, plus geographical controls and controls for industry sector; a vector of parameters to be estimated; and the error term, which assumes .

(1)

This is estimated by ordinary least squares (OLS) and then the difference between average men’s and women’s log wages (the wage gap) is decomposed via the Oaxaca–Blinder decomposition (Oaxaca, 1973):

(2) The first term is the part of the gap ‘explained’ by gender differences between characteristics (the endowment or productivity gap) and the second term accounts for any differing returns between men and women. Since this part is unexplained by the observable differences in

(9)

9

endowments, it is often called discrimination, although any effects from unobserved characteristics, such as ability not captured by education, would also be included.

However, a selectivity correction must be made because wages (or earnings) are only observed for those in employment statuses formal, informal employee and informal self-employed: it is a truncated wage sample. The expected value of the observed wage is:

where each individual n is in work status j if . Additionally, since participation in each employment type is not random given each individual’s characteristics – that is work status is endogenous – thus and there is selection bias in the average observed

wage. These selection biases mean that wage ‘offers’ (potential wages) – what an individual could earn, also based on how unobserved characteristics they have are valued in each employment status – do not equal the observed wages, and hence any calculation of differences in wages and decompositions thereof would be biased.

Stage 1:

A four-way selection between formal work, informal employees, informal self-employment and other is modelled via a multinomial logistic regression, estimated separately for men and women.2 A necessary assumption is the independence of irrelevant alternatives (IIA), such that an additional unchosen employment status is not relevant to the selection made.

If is the indirect utility associated with being in employment status j for individual n (modelled on , personal characteristics affecting employment status plus geographical controls, with error term ), then employment status j is observed when

. For a valid identifying exclusion restriction, the characteristics must

contain variables not in that can determine selection into employment status but not wages themselves, i.e. that are orthogonal. Following Ben Yahmed (2013), I include household size, number of young children, marital status and whether anyone else works formally in the household – plus I add whether an individual is head of the household to help differentiate selection for the informal self-employed – in to allow for supply-side characteristics. The sub-regional and urban area controls allow for the demand side. Under the IIA assumption that the residuals are independent and identically Gumbel distributed, the probability that individual n is observed in employment status j is given by:

(3) The model can be estimated via maximum likelihood for the log-likelihood function where is a dummy indicator taking 1 if individual n is in employment status j and 0 otherwise. The predicted overall probabilities, or proportions, of men and women in each employment type are calculated, .

(10)

10

Since this is a multinomial logisitic model, the selectivity correction required as an additional explanatory variable in the wage equation to control for the expected error is not the simple Inverse Mills Ratio version from the binary logistic model (Neuman & Oaxaca, 2004). Instead, following the improved Dubin–McFadden approach outlined in Bourguignon, Fournier & Gurgand (2007) and used by Ben Yahmed (2013) among others, which makes a normalised linearity assumption on the conditional mean of the residuals, the selectivity correction function for individual n in an observed employment type j is (with k = formal, informal emp, informal self, other):

(4)

where is the inverse normal of the cumulative of the Gumbel distribution of ; ;

; is the correlation between the errors in unadjusted wages, , and in selection, ;

and is the standard deviation of the unadjusted wage errors, , given and .

Stage 2:

Given the selection bias correction outlined above, the revised Mincerean wage equation to be estimated separately for men and women in formal, informal employee and informal self-employment status is:3

(5)

with the new error term. In addition to producing estimates for , , , ,

and , one set of coefficients per model, the terms necessary to calculate each

observation’s selectivity correction function are also estimated. This

function is formed of four variables (one related to each employment status) and their corresponding coefficients ( , , and from table 7).

Stage 3:

The separate logmeans are then given by:

Because of the inclusion of the selection correction functions, these errors now have mean zero and hence OLS estimation is consistent and also the following wage gap decompositions hold for each employment type in turn:

3_{Marc Gurgand and Martin Fournier’s selmlog command in Stata combines the multinomial logistic equation}

(11)

11

(6) Following Brown, Moon & Zoloth (1980), proportions across the three employment types with observed wages can be used to calculate the overall male and female means:

Brown, Moon & Zoloth use this decomposition as an early form of selectivity correction, but here I ascertain the importance of the observed distribution of men and women in employment types in addition to adjusting for unobserved differences driving the distribution via selection. Assuming that the male wage structure is the standard if there were no discrimination – a justifiable assumption based on Appleton, Hoddinott & Krishnan (1999)’s finding that it is female discrimination rather than male nepotism affecting wages in Uganda – the adjusted overall Oaxaca–Blinder wage gap decomposition is:

In addition to the usual explained and unexplained parts of the decomposition, the final six terms determine the gendered differential selection and proportion effects on the wage gap – a measure of the difference in average selection bias. Then, by rearranging terms:

for , s (7)

Each summation term gives the part of the wage gap decomposition that would disappear if there were no difference between men and women for each component that varies (g = m, w): (i) , proportions in each employment type; (ii) , average observable characteristics; (iii) , returns to endowments; and (iv) , selectivity corrections.

(12)

12

4. Data

The main data source is the third wave (2011–2012) from the Uganda National Panel household survey (UNPS). This was conducted by the Uganda Bureau of Statistics on a nationally representative sample of households, with the support of the World Bank. A total of 2,835 households (covering 16,139 individuals) in 322 enumeration areas were surveyed under a two-stage cluster sampling design. Considering only individuals aged 14 to 80 who were regular members of the households, not the sick, disabled nor those currently studying, and making sure that all basic personal information was provided, such as education and marital status, reduced the sample for analysis to a total of 5,717 individuals – 3,137 women and 2,580 men.4 The Uganda official working age is 14 to 64, but also counting older working people maximizes the data and allows those in work out of necessity in retirement (the segmented model for informality) to appear in the analysis.5

The national panel is a multi-purpose survey with a number of sub-sections; this research uses the labour force and household enterprise sections, plus those for education and basic individual and household characteristics. I consider the first (main) job or the first (main) household enterprise listed. The survey provides responses on employee benefits and contractual situation, and taxation status for household enterprises. By law, employees in Uganda are entitled to 21 days of paid annual leave after six months of service, sick leave after one month and must have a written contract within twelve weeks.6 There is effectively no legal minimum wage since this was frozen at 6,000 Ugandan shillings per month in 1999. Due to the high prevalence of informality in Uganda and the differing time periods at which employment benefits legally apply, I determine a definition of formality whereby meeting any criteria qualifies, enabling me to establish whether each individual is working formally, informally (either employees or self-employed) or belongs to the alternative other employment status (table 1). Under these criteria, of the total women’s sample 5% are formal, 5% informal employees and 13% informal self-employed; of the men’s sample 9% are formal, 14% informal employees and 14% informal self-employed.

Employee wages are considered monthly to match the self-employment earnings – monthly is also the most prevalent time period for which wages are provided.7 Thus the amount of time worked is subsumed into the monthly wage figure, which includes both pay/earnings and effort expended or, alternatively, any underemployment – which in Africa is on average twice as common for women than for men (Kolev & Sirven, 2010). Much of the literature on informal employment uses hourly wages for analysis because informal work is often only part-time, although Gunther & Launov (2012) feel monthly wages are a better reflection of informal earning opportunities and Nordman, Rakotomanana & Robilliard

4_{Simple pairwise deletion of observations was the only feasible solution to missing data but if those individuals}

for whom full data was not available were not random then this could introduce sample bias.

5_{A Wald test of all equal coefficients against a model using the 14–64 age range instead shows equality cannot}

be rejected at any significance level for any of the twelve permutations of model and base category to consider.

6_{The Ugandan Employment Act, 2006 (Act No. 6).} 7

Hourly wage data are only available in the sample for three individuals; the figures provided in tables 2 and 3 for reference are calculated backwards based on either actual hours, days and weeks worked per month or imputed average values for these.

(13)

13

(2010) consider both hourly and monthly wage gaps. There is a gender gap in time poverty rates, with women usually bearing the greatest burden of household chores in addition to market work (Saba Arbache, Filipiak & Kolev, 2010). Differing average hours worked between men and women, in formal and informal employment, are demonstrated in this Ugandan sample. In Madagascar, monthly wage gaps are larger than hourly ones, hence gendered differences in time use compound wage gaps (Nordman, Rakotomanana & Robilliard, 2010).

Table 1: Employment status determinants

Employment status Employee pay or enterprise earnings Remunerated labour force activity1

Main job in survey 2 Benefits/taxation

Formal Pay > 0 Employee Working for someone else for pay

Employer pension contributions, paid leave entitlement, medical benefits, income tax deducted/paid, a written contract or permanent pensionable status Earnings > 0 Managing non-agricultural household enterprise3 Employer, own-account worker, helping with household business or working on household farm1,2

Enterprise registered for VAT or income tax

Informal employee

Pay > 0 Employee Working for someone else for pay

No benefits, no pension, no income taxation and verbal contract only Informal self-employed Earnings > 04 Managing non-agricultural household enterprise3 Employer, own-account worker, helping with household business, working on household farm, apprentice or no response1,2

Enterprise registered for neither VAT nor income tax Other (non-wage-earning) No pay & no earnings (or earnings < 04)

None Working on household farm, helping with household business, own-account worker, employer, apprentice, other or no response

N/A

1_{If an individual has both positive pay and earnings then whichever is greater defines whether they are counted as an employee of self-employed.} In the case of dual employment, the main job in the survey is given as working for someone else for pay even though in this analysis they may be counted as self-employed if earnings are greater than pay.

2_{This question appears in the labour force section of the survey, not in the household enterprise section. Therefore for those managing a} household enterprise (the self-employed as defined in this analysis) there are some responses other than employer or own-account worker. 3_{Despite being described as non-agricultural, the enterprise can involve processing and/or selling crop output if this is a regular activity.} 4

Although managing a loss-making enterprise would usually be counted as informal self-employment, due to the functional form of the two-stage wage equation estimation, only strictly positive wages/earnings can be modelled and therefore loss-making enterprises are counted in the non-wage-earning employment status (unless that individual is also an employee with positive pay, in which case they are counted as an employee).

(14)

14

Table 2: Descriptive statistics by type of employment: Women (1) Formal (2) Informal employees (3) Informal self-employed (4) Other mean st. dev. mean st. dev. mean st. dev. mean st. dev. Age 35.84 (11.05) 30.14 (12.88) 39.16 (12.15) 37.36 (15.34) Married 0.58 (0.49) 0.29 (0.46) 0.67 (0.47) 0.67 (0.47) Head of household 0.32 (0.47) 0.26 (0.44) 0.38 (0.49) 0.22 (0.42) Household size 5.73 (3.10) 6.05 (3.97) 6.12 (2.79) 6.67 (3.15) Number of children < 1 in household 0.14 (0.35) 0.17 (0.38) 0.15 (0.37) 0.20 (0.43) Number of children < 5 in household 0.67 (0.84) 1.01 (1.04) 0.85 (0.85) 1.05 (1.02) Number of children < 14 in household 1.91 (1.65) 2.45 (2.26) 2.75 (1.80) 3.06 (2.10) Anyone else formal in household 0.38 (0.49) 0.16 (0.37) 0.09 (0.28) 0.08 (0.27) Illiterate – can neither read nor write 0.09 (0.29) 0.27 (0.45) 0.30 (0.46) 0.42 (0.49) Years of schooling1 10.70 (3.98) 6.26 (4.30) 5.19 (3.65) 4.39 (3.70) Some basic schooling 0.17 (0.38) 0.43 (0.50) 0.57 (0.50) 0.58 (0.49) Completed primary 0.45 (0.50) 0.31 (0.46) 0.21 (0.41) 0.14 (0.35) Completed secondary 0.12 (0.32) 0.01 (0.12) 0.01 (0.07) 0.01 (0.08) Completed university 0.12 (0.32) 0.01 (0.12) 0.01 (0.07) 0.01 (0.08) Largest ethnic group: Baganda 0.30 (0.46) 0.35 (0.48) 0.24 (0.43) 0.15 (0.35) Urban 0.50 (0.50) 0.46 (0.50) 0.28 (0.45) 0.17 (0.37) Sub-region: Kampala 0.17 (0.37) 0.26 (0.44) 0.08 (0.27) 0.05 (0.22) Central 1 0.17 (0.38) 0.22 (0.42) 0.14 (0.35) 0.10 (0.30) Central 2 0.12 (0.32) 0.09 (0.29) 0.15 (0.35) 0.10 (0.29) East Central 0.10 (0.30) 0.03 (0.16) 0.04 (0.20) 0.11 (0.31) Eastern 0.09 (0.29) 0.02 (0.14) 0.06 (0.24) 0.16 (0.37) Mid-north 0.08 (0.27) 0.06 (0.23) 0.12 (0.33) 0.10 (0.30) North East 0.03 (0.18) 0.06 (0.23) 0.07 (0.25) 0.06 (0.24) West Nile 0.04 (0.20) 0.09 (0.29) 0.14 (0.35) 0.10 (0.30) Mid-west 0.08 (0.27) 0.06 (0.24) 0.07 (0.26) 0.09 (0.29) South West 0.13 (0.33) 0.12 (0.32) 0.12 (0.33) 0.13 (0.33) Industry: 2

Agriculture & Fishing 0.04 (0.20) 0.24 (0.43) 0.04 (0.18) Manufacturing 0.04 (0.20) 0.03 (0.18) 0.13 (0.34) Mining & Construction 0.01 (0.08) 0.01 (0.08) 0.01 (0.09) Retail & Hospitality 0.10 (0.30) 0.17 (0.37) 0.40 (0.49) Transport, Comms & Utilities 0.01 (0.08) 0.01 (0.08) 0.01 (0.10) Finance & Real Estate 0.01 (0.12) 0.02 (0.14) 0.02 (0.15) Public Admin 0.04 (0.20) 0.00 (0.00) 0.02 (0.14) Education 0.44 (0.50) 0.12 (0.32) 0.00 (0.00) Health & Social Work 0.11 (0.32) 0.05 (0.22) 0.01 (0.10) Community 0.10 (0.31) 0.21 (0.41) 0.03 (0.17) Other2 0.05 (0.22) 0.15 (0.35) 0.00 (0.05) Employees 0.84 (0.37)

Average days worked in week3 5.17 (1.20) 5.47 (1.55) Average hours worked in week3 44.85 (16.67) 48.05 (24.97) Hourly wages3 (Ug Shs) 2,404 (4,228) 1,040 (4,364)

Monthly wages/earnings4 (Ug Shs) 302,364 (308,424) 100,821 (86,727) 95,618 (186,908)

N 144 144 396 2,453

Source: Author’s calculations based on the 2011/12 UNPS.

1_{Years of schooling is based on grades completed. Post-primary (3 years) and post-lower-secondary (2 years) education is calculated according to figures in} SACMEQ (n.d.); university education is estimated at 4 years (the mean of 3 and 5 years).

2_{About a third of the self-employed do not have an industry sector. The other category encompasses private household employers and extra-terrestrial organisations.} 3_{Hourly wages and hours/days worked are only available for employees, not the self-employed.}

4_{Monthly self-employment earnings are calculated as: gross earnings minus expenses for raw materials, non-family hired employee wages and other operating} expenses, and assumed to be split equally if two individuals run the business jointly, with any other household members not running the business assumed to be unpaid. Monthly employee wages include both cash and in-kind payments. If wages are given in terms of hours, days or weeks and the number worked is not provided, then means within gender and employment type are used to impute the monthly wage.

(15)

15

Table 3: Descriptive statistics by type of employment: Men (1) Formal (2) Informal employees (3) Informal self-employed (4) Other mean st. dev. mean st. dev. mean st. dev. mean st. dev. Age 39.06 (11.33) 32.83 (12.67) 41.02 (12.79) 38.51 (16.74) Married 0.79 (0.41) 0.57 (0.50) 0.88 (0.32) 0.67 (0.47) Head of household 0.78 (0.41) 0.58 (0.49) 0.90 (0.31) 0.64 (0.48) Household size 6.19 (3.13) 5.92 (3.44) 6.16 (2.99) 6.83 (3.05) Number of children < 1 in household 0.16 (0.39) 0.17 (0.38) 0.20 (0.43) 0.18 (0.40) Number of children < 5 in household 0.90 (0.92) 0.96 (0.95) 1.08 (0.95) 0.95 (0.98) Number of children < 14 in household 2.44 (1.88) 2.38 (2.05) 2.89 (1.90) 2.95 (2.01) Anyone else formal in household 0.19 (0.39) 0.07 (0.26) 0.04 (0.19) 0.05 (0.21) Illiterate – can neither read nor write 0.05 (0.21) 0.19 (0.39) 0.14 (0.35) 0.21 (0.41) Years of schooling1 10.47 (3.97) 6.36 (3.60) 6.77 (3.70) 5.97 (3.61) Some basic schooling 0.22 (0.42) 0.56 (0.50) 0.59 (0.49) 0.64 (0.48) Completed primary 0.51 (0.50) 0.31 (0.46) 0.28 (0.45) 0.22 (0.41) Completed secondary 0.12 (0.32) 0.00 (0.00) 0.02 (0.15) 0.01 (0.11) Completed university 0.12 (0.32) 0.00 (0.00) 0.02 (0.15) 0.01 (0.11) Largest ethnic group: Baganda 0.21 (0.41) 0.25 (0.43) 0.22 (0.41) 0.13 (0.34) Urban 0.42 (0.49) 0.35 (0.48) 0.20 (0.40) 0.14 (0.35) Sub-region: Kampala 0.17 (0.38) 0.15 (0.36) 0.06 (0.24) 0.03 (0.18) Central 1 0.11 (0.31) 0.15 (0.36) 0.13 (0.34) 0.09 (0.29) Central 2 0.09 (0.29) 0.10 (0.30) 0.13 (0.34) 0.08 (0.28) East Central 0.10 (0.30) 0.08 (0.27) 0.10 (0.29) 0.11 (0.31) Eastern 0.08 (0.27) 0.10 (0.30) 0.12 (0.32) 0.15 (0.36) Mid-north 0.13 (0.33) 0.04 (0.19) 0.11 (0.32) 0.11 (0.32) North East 0.01 (0.11) 0.02 (0.13) 0.03 (0.16) 0.07 (0.25) West Nile 0.08 (0.27) 0.08 (0.27) 0.10 (0.30) 0.12 (0.32) Mid-west 0.12 (0.32) 0.11 (0.31) 0.09 (0.29) 0.11 (0.31) South West 0.12 (0.32) 0.18 (0.38) 0.13 (0.34) 0.12 (0.32) Industry: 2

Agriculture & Fishing 0.07 (0.25) 0.26 (0.44) 0.07 (0.25) Manufacturing 0.09 (0.28) 0.05 (0.23) 0.10 (0.30) Mining & Construction 0.04 (0.20) 0.22 (0.42) 0.03 (0.17) Retail & Hospitality 0.13 (0.34) 0.13 (0.33) 0.34 (0.47) Transport, Comms & Utilities 0.08 (0.27) 0.14 (0.35) 0.07 (0.26) Finance & Real Estate 0.01 (0.09) 0.00 (0.05) 0.01 (0.09) Public Admin 0.13 (0.33) 0.01 (0.09) 0.02 (0.15) Education 0.26 (0.44) 0.03 (0.18) 0.01 (0.11) Health & Social Work 0.05 (0.21) 0.01 (0.09) 0.01 (0.09) Community 0.12 (0.33) 0.11 (0.32) 0.06 (0.24) Other2 0.02 (0.16) 0.03 (0.16) 0.00 (0.05) Employees 0.87 (0.33)

Average days worked in week3 5.56 (1.23) 5.45 (1.57) Average hours worked in week3 50.15 (17.85) 46.25 (21.10) Hourly wages3 (Ug Shs) 2,869 (5,152) 1,622 (2,695)

Monthly wages/earnings4 (Ug Shs) 374,711 (367,872) 227,974 (281,446) 181,073 (327,975)

N 243 370 356 1,611

Source: Author’s calculations based on the 2011/12 UNPS.

1_{Years of schooling is based on grades completed. Post-primary (3 years) and post-lower-secondary (2 years) education is calculated according to figures in} SACMEQ (n.d.); university education is estimated at 4 years (the mean of 3 and 5 years).

2_{About a third of the self-employed do not have an industry sector. The other category encompasses private household employers and extra-terrestrial organisations.} 3_{Hourly wages and hours/days worked are only available for employees, not the self-employed.}

4_{Monthly self-employment earnings are calculated as: gross earnings minus expenses for raw materials, non-family hired employee wages and other operating} expenses, and assumed to be split equally if two individuals run the business jointly, with any other household members not running the business assumed to be unpaid. Monthly employee wages include both cash and in-kind payments. If wages are given in terms of hours, days or weeks and the number worked is not provided, then means within gender and employment type are used to impute the monthly wage.

(16)

16

Tables 2 and 3 provide demographic, educational, work and household characteristics of women and men in the sample according to the four employment statuses. Both for men and women, those working formally are older than those working informally as employees, but the informal self-employed are the oldest on average of all employment statuses (including older than the non-wage-earning) and there are similar patterns for proportions married and head of the household. Women working formally have on average a smaller household size and with fewer children of any age living in the household compared to all other employment statuses. In contrast, men working formally come from larger households, on average, than those working informally, although still with fewer young children. For men, it is the informal employee status with the smallest average household size and lowest number of children overall. The UNPS only contains data on households not on families, and thus household rather than family attributes are considered.

Having another person working formally in the household occurs in a larger proportion of cases for formal women (38%) than for men (19%), suggesting that solo formality for women is less feasible, or that there is assortative matching for married formal women. As would be expected, having completed more schooling, particularly secondary and university education, is more prevalent in those working formally for both men and women. Formal women have on average slightly more years of schooling than formal men (10.70 compared to 10.47 years) despite the fact that over the whole sample women have on average considerably fewer years of schooling than men (4.87 compared to 6.56 years) and are more likely to be illiterate (38% compared to 18%). The opposite is true for both employee and self-employed informal women compared to men – informal women have on average completed fewer years of education than informal men, particularly in the case of the informally self-employed.

Uganda is a very ethnically diverse country, with the largest ethnic group in the sample, the Baganda, accounting for just 17% of individuals and the next largest ethnic group accounting for less than 10%. A disproportionately large 30% of formal women are of Baganda ethnicity and 21% of formal men. This seems most likely due to the fact that the Baganda kingdom covers much of the Central region of Uganda, including the capital, Kampala. A greater proportion of formal men are to be found in Kampala and for women, larger proportions work formally and as informal employees in Kampala and the Central 1 sub-region.

Considering work-related characteristics, for both men and women the vast majority of formal workers are employees rather than self-employed. However, there is a noticeable gender difference for informal workers, with only 27% of informal women employees compared to 51% of informal men. Education is the largest sector for formal men (26%) as well as for formal women (44%), but the figure is considerably higher for formal women. The retail and hospitality sector is where the largest single group of both informally self-employed women (40%) and men (34%) work. In contrast, for informal employees there are gender differences, with agriculture and fishing (26%) and mining and construction (22%) significant sectors for men but community (21%) and retail and hospitality (17%) also large groups alongside agriculture and fishing (24%) for women. The employment profiles of men and women do appear to differ, and this is seen in the figures for how much individuals work and how much they are paid.

(17)

17

Kernel Density Plots of Log Wage Distributions

Figure 2A: Log wages of women and men Figure 2B: Log wages of formal and informal (employees and the self-employed)

Figure 3A: Log wages of women: formal and informal (employees and the self-employed)

Figure 3B: Log wages of men: formal and informal (employees and the self-employed)

Figure 4A: Log wages of formal: women and men Figure 4B: Log wages of informal employees: women and men

Figure 4C: Log wages informal self-employed: women and men

(18)

18

Women work more hours each week on average if they are informal employees compared to formal employees (48.05 compared to 44.85 hours/week) but earn considerably less – average women informal employees’ monthly wages are 100,821 compared to 302,364 Ugandan shillings for formal women. In contrast, men work less if they are informal rather than formal employees (46.25 compared to 50.15 hours/week) and, although they still earn less on average informally compared to formally, with average men informal employees’ monthly wages of 227,974 compared to 374,711 Ugandan shillings for formal men, this is 61% of formal wages, whereas for women informal employees’ wages are 33% of formal. Informal self-employment earnings are lower again than informal employee earnings on average – 95% in the case of women and 79% for men. Working informally as a woman in Uganda seems to be a particular economic disadvantage, although proportionally the disadvantage is smaller for informal self-employed compared to informal employee women.

The average wage across all men in the sample is higher than for women (247,541 compared to 136,235 Ugandan shillings), a difference which can also be seen in the raw gap for (log) wage distributions in figure 2A.8,9 As shown in figures 2B, 3A and 3B, the distribution of formal (log) wages lies to the right of – is higher than – that of informal employee (log) wages, which itself lies to the right of informal self-employed (log) wages for men, women and both combined. However, the overlaps between formal and informal wage distributions demonstrate that not all informality is economically inferior to formality. Significant differences in mean and distribution between the two types of informal wages inform my decision to split informal workers into employees and the self-employed.

Looking at the formal (figure 4A) and informal (figures 4B and 4C) (log) wages separately, the men’s and women’s distributions are very similar for formal but clearly demonstrate a gender difference for both informal employees and the informal self-employed.10 In the case of the self-employed there could be unequal enterprise attributes and potentially the effect of differential access to physical capital (Nordman, Rakotomanana & Robilliard, 2010). Focusing the analysis solely on employees, though, would not accurately estimate the size of the gender wage gap as well as incorrectly modelling gender differences in selection into informal employment.

5. Results

5.1. Selection model

The first step in the analysis is a multinomial logistic employment status selection (equation 3), estimated separately for men and women. In order to most efficiently model this from the data, survey sampling weights are not used (Deaton, 1997; Solon, Haider & Wooldridge, 2013). However, observations are clustered at the level of enumeration area (the primary sampling unit) to correctly adjust the standard errors. Much of the literature analysing gender

8_{The top and bottom 1% of the wage distribution are removed to eliminate potential measurement error outliers.} 9

The result of a Welch’s t-test shows that the gap is significant at all confidence levels and a Kolmogorov-Smirnov test similarly rejects the hypothesis that the distributions of male and female (log) wages are the same.

(19)

19

wage gaps does not look in detail at the selection models, but I believe it is instructive. With four employment statuses, the model of relative probabilities is complex so I also consider average adjusted probability (AAP) predictions. These demonstrate what the probabilities of the actual individuals in the sample would be for each employment status, adjusting one explanatory variable at a time with all else kept as is. The corresponding average marginal effects (AMEs) similarly measure the effect of marginal changes, or discrete changes for indicator variables. For continuous variables such as age and years of education, I also plot the probability predications and marginal effects at different values.

There are four key selection results to note. First, there is a gender difference in how education impacts on the likelihood of the two informal statuses (figures 7B and 7C). The adjusted probability of being an informal employee declines for men (a negative marginal effect) but for women it first increases slightly until about six years of education, which is nearly the end of primary school in the Ugandan education system. There are mixed effects of education on being informally self-employed. The increased probability up to ten or eleven years of schooling (around the end of lower secondary) is likely due to the increase against non-wage-earning and informal employee status (table 5, columns 1 and 4), then the decreased probability due to the decrease against formal employment (table 5, column 5). These effects are at different scales for men and women. On average, informal women are less educated than informal men – particularly the self-employed – despite the positive (for women) and negative (for men) education impacts (table 4, columns 2 and 7). This is an artifact of the lower overall base level of education for women.

Table 4: Multinomial logistic selection model – Average marginal effects

Women Men (1) (2) (3) (4) (5) (6) (7) (8) Other Informal self-employed Informal employees

Formal Other Informal self-employed Informal employees Formal Years of education -0.015*** 0.004*** -0.001 0.011*** -0.012*** 0.002 -0.008*** 0.019*** (0.002) (0.002) (0.001) (0.001) (0.003) (0.002) (0.002) (0.002) Age† -0.002*** 0.002*** -0.001* 0.001*** -0.002** 0.001* -0.001 0.002*** (0.001) (0.000) (0.000) (0.000) (0.001) (0.001) (0.001) (0.001) Married 0.076*** 0.015 -0.075*** -0.015* 0.027 0.039* -0.075*** 0.009 (0.019) (0.014) (0.012) (0.009) (0.031) (0.021) (0.026) (0.021) Head of the household -0.075*** 0.065*** -0.004 0.014 -0.067** 0.084*** -0.020 0.003

(0.020) (0.017) (0.012) (0.010) (0.032) (0.020) (0.024) (0.020) Household size 0.010*** -0.002 -0.005*** -0.003** 0.023*** -0.006** -0.014*** -0.003

(0.003) (0.002) (0.002) (0.001) (0.004) (0.003) (0.004) (0.002) Anyone else formal in -0.060** -0.017 0.016 0.062*** -0.029 -0.037 0.006 0.061**

the household (0.026) (0.019) (0.016) (0.017) (0.040) (0.028) (0.027) (0.026) Number of children -0.005 -0.005 0.014*** -0.004 -0.048*** 0.016* 0.029*** 0.003 under 5 in the hh (0.009) (0.007) (0.005) (0.005) (0.012) (0.009) (0.010) (0.008) Urban -0.095*** 0.040** 0.040** 0.015 -0.158*** 0.001 0.141*** 0.015 (0.024) (0.019) (0.016) (0.010) (0.025) (0.022) (0.025) (0.013) N 3,137 2,580 †

Since the marginal effect of Age2 actually comes from the marginal effect of Age, these are jointly estimated together and hence there is no separate Age2 entry. Robust standard errors in parentheses

(20)

20

Selection Model: Average Adjusted Probabilities

Figure 5A: Women and years of education

Figure 5B: Men and years of education

Figure 6A: Women and age

Figure 6B: Men and age

Since this is a non-linear model, interpreting the values of the coefficients for log odds is difficult, and therefore odds ratios results with a relative risk ratio (RRR) interpretation are presented in table 5.11 The stronger positive impacts of years of education on formality and less negative impacts on informal employee status for women compared to men in table 5 could be due to a greater possibility of women quitting work (for pregnancy or other family commitments) and hence requiring more human capital to compensate for this to employers. As expected from the descriptive statistics, additional years of education overall increases the probability of working formally and decrease the probability of being a non-wage-earner (table 4 and figures 5A and 5B).

The second key result is that the effect of young children in the household (a count of the number aged under five) is both stronger and broader for men than for women. This is unexpected from the literature. Young children increase the probability of men being in either type of informal work and decrease the probability of non-wage-earning (table 4) – perhaps due to the traditional male role of household breadwinner. According to OECD (2015, p. 49), in Uganda, boys ‘delay marriage because they fear responsibility of providing for the family’. In contrast, from table 4, column 3, we can see that more young children in the household only increases the probability of informal employee status for women, although this is a strong and significant result with respect to all three alternative employment statuses (table 5, panel A, columns 2, 4 and 6). The reason is unclear.

11

Any RRR coefficient greater than one means a positive impact (corresponding to a positive coefficient for the log odds) and vice versa there is a negative impact for an RRR coefficient less than one (corresponding to a negative coefficient for the log odds); the further above/below unity the larger the effect.

(21)

21

Selection Model: Average Marginal Effects

Figure 7A: Women and men – formal

by years of education

Figure 7B: Women and men – informal employees by years of education

Figure 7C: Women and men – informal self-employed by years of education

Figure 7D:

Women and men – non-wage-earning by years of education

Figure 8A: Women and men –

formal by age

Figure 8B: Women and men – informal employees by age

Figure 8C: Women and men – informal

self-employed by age

Figure 8D: Women and men – non-wage-earning by age

(22)

22

The third key result is that, despite matching strong positive effects on the likelihood of formality from another formal worker in the household, this effect is much stronger against other employment statuses for women than men (table 5). There is also an overall negative impact on non-wage-earning for women (table 4, column 1). Ben Yahmed (2013) postulates that another household worker with the security of a formal job might make formal employment less valuable to women – but that does not appear to hold here. Instead, perhaps the effect is a form of the psychosocial assistance described by Campos et al (2015), or alternatively a sign of assortative matching for married women.

The fourth and final key result is the effect of an urban setting, which increases the likelihood of wage-earning work in general (table 4, columns 1 and 5) and informal work in particular. For women the positive effect is across both informal self-employment and being an informal employee, but for men the effect is significantly larger and only for informal employees. From table 5, panel B, columns 4 and 6, we can see that living in an urban environment increases the relative risk of informal employee status for men with respect to both formal work and informal self-employment. There are no such effects in the women’s model – perhaps due to Saba Arbache, Filipiak & Kolev (2010)’s finding that there is a stronger gender-based division of labour in rural areas in African nations such as Ethiopia.

There are other minor points in the selection models. The effect of age (proxying for experience) averages out to near zero (table 4) but this is due to positive effects of being older up to age 40-something (for formal employment and informal self-employment) followed by negative effects for increased age after this point. These effects are mirrored in reverse on the probability of being a non-wage-earner. Hence there is convexity in figures 5C and 5D from the Age2 variable in table 5. The picture for informal employees differs between the genders, with age always having a negative effect for women on the likelihood of being an informal employee. Holding all else constant, being older increases the likelihood of women working as informal self-employed with respect to being an informal employee (table 5, panel A, column 4). A gender difference is observed for the effects of marriage: an overall positive effect for women on the probability of non-wage-earning and a negative effect on formal employment (both insignificant for men), whereas for men marriage has a negative effect on informal self-employment (insignificant for women). Despite this, there is actually a large increased relative risk of formal work with respect to informal employees for women due to being married. Since the majority of formal work is that of employees, this suggests that marriage is a strong indicator for formality among employee women.

Being head of the household unsurprisingly decreases the likelihood of non-wage-earner status for both men and women, but it also increases the likelihood of being informally self-employed without any significant impact on the other employment types. The negative effect is stronger for women, the positive effect stronger for men. A larger household increases the probability of being a non-wage-earner – possibly as there is more chance of someone else being a wage-earner in a larger household – and decreases the probability of all types of wage-earning employment. However, this negative effect is greatest on men being informal employees. Looking at table 5, panel A, column 4, we can see that larger household size also increases the relative risk of women being informally self-employed with respect to being informal employees. Along with age and being head of the household, household size is a significant explanatory factor differentiating the two types of informal work for women.

(23)

23

Table 5: Multinomial logistic selection model – Relative risk ratios (odds ratios)

(1) (2) (3) (4) (5) (6) InfSelf† wrt Other InfEmp† wrt Other Formal wrt Other InfEmp wrt InfSelf Formal wrt InfSelf Formal wrt InfEmp Panel A: Women Years of education 1.070*** 1.024 1.430*** 0.957 1.337*** 1.396*** (0.018) (0.034) (0.056) (0.035) (0.053) (0.069) Age 1.209*** 1.050 1.322*** 0.869*** 1.094 1.258*** (0.032) (0.038) (0.084) (0.038) (0.072) (0.088) Age2 0.998*** 0.999** 0.997*** 1.001** 0.999 0.998*** (0.000) (0.000) (0.001) (0.001) (0.001) (0.001) Married 0.988 0.166*** 0.551** 0.168*** 0.557** 3.326*** (0.146) (0.040) (0.138) (0.044) (0.142) (1.111) Household head 1.847*** 1.035 1.719* 0.561* 0.931 1.660 (0.256) (0.327) (0.481) (0.189) (0.276) (0.652) Household size 0.964 0.861*** 0.895*** 0.893** 0.929* 1.040 (0.023) (0.045) (0.037) (0.048) (0.039) (0.056) Anyone else working 0.985 1.629 4.105*** 1.654 4.166*** 2.519**

formally in the household (0.196) (0.532) (1.138) (0.593) (1.345) (1.010) Number of children under 5 0.966 1.409*** 0.903 1.459*** 0.935 0.641***

in the household (0.069) (0.166) (0.126) (0.191) (0.131) (0.110) Urban 1.583*** 2.681*** 1.826** 1.694 1.154 0.681 (0.252) (0.811) (0.479) (0.549) (0.328) (0.262) Constant 0.003*** 0.246** 0.000*** 84.676*** 0.009*** 0.000*** (0.002) (0.173) (0.000) (79.497) (0.014) (0.000) Panel B: Men Years of education 1.044** 0.967* 1.311*** 0.926*** 1.255*** 1.355*** (0.019) (0.017) (0.034) (0.021) (0.038) (0.040) Age 1.130*** 1.086*** 1.242*** 0.961 1.099** 1.144*** (0.034) (0.031) (0.049) (0.038) (0.053) (0.055) Age2 0.999*** 0.999*** 0.998*** 1.000 0.999** 0.999** (0.000) (0.000) (0.000) (0.000) (0.001) (0.001) Married 1.309 0.561*** 1.064 0.429*** 0.813 1.897* (0.288) (0.116) (0.338) (0.117) (0.285) (0.687) Head of the household 2.482*** 0.975 1.201 0.393*** 0.484* 1.233

(0.649) (0.209) (0.357) (0.121) (0.182) (0.387) Household size 0.914*** 0.862*** 0.917*** 0.944 1.004 1.064

(0.025) (0.029) (0.031) (0.038) (0.042) (0.046) Anyone else working 0.789 1.117 2.008** 1.416 2.545** 1.798**

formally in the household (0.249) (0.282) (0.556) (0.519) (0.972) (0.530) Number of children under 5 1.241*** 1.373*** 1.158 1.107 0.933 0.843

in the household (0.099) (0.123) (0.132) (0.119) (0.115) (0.116) Urban 1.384* 3.226*** 1.664*** 2.331*** 1.202 0.516***

(0.258) (0.541) (0.302) (0.571) (0.301) (0.113) Constant 0.011*** 0.367* 0.000*** 33.210*** 0.045*** 0.001***

(0.006) (0.214) (0.000) (25.321) (0.043) (0.001)

Sub-region dummies yes yes yes yes yes yes

Clustering at community level yes yes yes yes yes yes

N Women: 3,137 Men: 2,580 Log likelihood -1895 -2365 Wald chi2 725.6 751.4 McFadden’s pseudo R2 0.18 0.14 †

InfSelf stands for Informal self-employment and InfEmp stands for Informal employees Robust standard errors in parentheses

(24)

24

Wald tests on each of the individual explanatory variables in the selection model provide evidence that all the explanatory variables are individually significant at the 99% confidence level in at least one selection model (or for the sub-regional dummy variables, significant as a group).12 These results, together with Wald chi2 tests of both models as a whole, demonstrate that the models do not contain insignificant variables and do partially predict employment status. Another Wald test, for the possibility of combining any two of the employment outcome statuses in each model, rejects the null that any could be combined at all confidence levels.13 The models as specified are satisfactory.14

One of the assumptions required for a multinomial logistic model is the independence of irrelevant alternatives (IIA), although Long & Freese (2006) discuss shortcomings of various IIA tests. Using a suest-based Hausman-type test that accounts for clustering, the results are mixed: of the forty-eight different permutations of model, base category and removed category to consider, joint equality of coefficients in eight such combinations can be rejected at the 99% confidence level. Although these results mean that IIA is violated, given that the estimated coefficients are comparable and none were found to be insignificant by the Wald tests – plus the standard model from the literature is a single multinomial logistic selection model – this model was kept.15 Perhaps a more accurate model is nested: first a binary selection of whether to be in the wage-earning workforce and then a subsequent selection of formality vs the two types informality conditional on the first selection.

For all significant marginal effects in the women’s model (except age, which averages to near insignificance due to differing effects at different ages), selecting into being a non-wage earner has the opposite sign of the effects for selecting into any formal or informal employment (table 4). This suggests that for women the biggest difference is whether selecting into wage-earning or not, which conforms to the literature.

Given the differences of scale as well as effects, sorting into the three employment statuses is a gendered process.16 Although unseen preferences will form a large part of the selection, the explanatory variables modelled here demonstrate that men and women are selected differentially into employment statuses.

5.2. SIGI discrimination

The secondary data source used in this research is the OECD Development Centre’s Social Institutions and Gender Index (SIGI), which measures discriminatory social norms. Indicators cover legislative positions, actual practices and attitudes in five areas: (i) discriminatory family code, (ii) restricted physical integrity, (iii) son bias, (iv) restricted access to resources

12_{This holds in both models at the 95% confidence level. It is essential for the wage selectivity corrections that}

the exclusion restrictions are significant variables in the selection models.

13_{Many post-estimation tests are from the SPost13 set of user-written commands: J. Scott Long & Jeremy}

Freese (2014), Regression Models for Categorical Dependent Variables Using Stata, 3rd Edition, Stata Press.

14_{The Fagerland–Hosmer–Bofin multinomial version of the Hosmer–Lemeshow goodness-of-fit tests for each}

model and base category results in insignificant test statistics at the 96% confidence level.

15_{According to Bourguignon, Fournier & Gurgand (2007), the Dubin–McFadden selectivity adjustment used for}

the wage analysis still provides a fairly good correction even if the IIA hypothesis is violated.

16_{This is confirmed by suest-based Wald tests showing that the coefficients for the men’s and women’s models}

Gender wage gaps, informality and discriminatory social norms : empirical evidence from Uganda