Self-employment and Conflict in Colombia

Abstract

Many Colombians are confronted with the ongoing conflict that influences their decision making in everyday life, including their behavior in labor markets. This study focuses on the impact of violent conflict on self-employment, enlarging the usual determinants with a set of conflict variables. Our estimation strategy compares three different estimates: one from fixed-effects panel data (OLS-FE), estimates using lagged conflict indicators instead of contemporaneous regressors, and instrumental variables (IV-FE) estimates. Our results show that a one standard deviation increase in net displacement rates increases the rate of self-employment by about 7 percent points. Dividing the self-employed into different sectors (services and agriculture), we find that net displacement increases self-employment in the services sector but has no effect in agriculture that is affected by attacks by rebel and paramilitary groups, instead. Looking at the income of self-employed individuals, an influx of displaced reduces sharply hourly income in the self-employment sector.

Keywords

self-employment civil conflict rural labor markets Colombia

Does violent conflict impact the share of (informal) self-employed workers in developing countries? While self-employment and war are relevant topics in developing countries, there is a scarcity of studies that specifically deal with the effects of violent conflict on self-employment. For this reason, our aim is to bridge this research gap by analyzing the effect of violent conflict on the probability to be self-employed.

We investigate this topic focusing on rural Colombia, an area with all the ingredients for this kind of study: on one hand, it has suffered a violent conflict for more than forty years, and, on the other hand, the share of self-employed increased from 20 percent to 30 percent over the last twenty years. Last, but not least, detailed data sets at the micro level are available for multiple years, allowing us to use panel data estimation techniques.

One particular characteristic of the Colombian conflict is the high rate of internal displacement. Between 1998 and 2008, 4.2 million people were internally displaced, representing about 10 percent of the population (Calderón and Ibáñez 2009). The main reason for internal displacement is threats by irregular armed groups, either by paramilitaries or by rebel groups like the Revolutionary Armed Forces of Colombia (FARC) and the National Liberation Army (ELN). These actors exercise violence against the civilian population in order to maintain control over the territory and to expulse farmers from their land using it subsequently for the cultivation of illegal crops. In 2009, 89 percent of displacement in Colombia was individual displacement. Displacement mainly occurs from rural regions to mid-sized and big cities. On average, displaced families are worse off than nondisplaced families in a variety of socioeconomic status indicators. Additionally, the income of 98.6 percent of displaced families is below the poverty line, and 82.6 percent have an income below the indigence line (IDMC 2009).

Using different types of estimates to control for the potential endogeneity of conflict indicators, our main results indicate that self-employment rates increase in response to an inflow of displaced individuals and in response to rebel group attacks. Thus, conflict not only impacts the self-employed living in directly affected communities (via attacks) but also on self-employment shares in communities that experience the consequences of conflict but not the conflict itself (via the reception of displaced individuals).

Additionally, we estimated the impact of conflict separately for being self-employed in the services or the agricultural sector. By doing so, we find that self-employment in services is increasing in municipalities that are net receivers of internally displaced people (this is to say, with a net displacement greater than zero). Net displacement does not have an impact on self-employment in agriculture, but attacks by irregular armed groups do: municipalities that experience more attacks display a greater share of self-employed individuals in the agricultural sector. Finally, we also analyze the effect of income from self-employment concluding that income of self-employed is affected by an influx of displaced individuals, while income from wage employment is not.

The results of our study contribute to two strands of literature: the self-employment and the conflict literatures. The literature in both areas is growing rapidly. Most research on self-employment focuses on the impact of earnings, access to capital as well as individual characteristics like gender, education, labor market experience, and attitude to risk, but none of these studies investigate the effect of conflict on self-employment.

The conflict literature concentrates mainly on the impact of conflict on poverty, education, migration, health, household welfare, and consumption (Justino and Verwimp 2006; Shemyakina 2006; Grun 2008; Ibáñez and Vélez 2008; Bundervoet, Verwimp, and Akresh 2009; Rodríguez and Sánchez 2009). Research on the effects of civil conflict on labor markets, especially at the micro level, is virtually nonexistent with only a few notable exceptions: Deininger (2003) detects that violent conflict leads to a reduction in investment of nonagricultural enterprises in Uganda. He uses two waves of panel data from Uganda but concentrates on a particular type of self-employment and does not control for the potential endogeneity of reported violence. Kondylis (2010) uses an instrumental variable strategy on longitudinal postconflict data from Bosnia-Herzegovina to identify the effect of displacement on labor-market outcomes. Her findings focus on postconflict “legacies,” while we focus on ongoing conflict effects. Finally, Calderón and Ibáñez (2009) use disaggregated conflict data to instrument for the share of forced migrants relocated in Colombian metropolitan areas. They find that an inflow of forced migrants leads to significant changes in employment outcomes, especially in the informal economy. We focus on the rural poor of Colombia: in our sample, informality rates are overwhelmingly high¹; thus, self-employment is a relevant topic for two reasons. First, it is an important aspect of entrepreneurship, which is the focus of this Special Issue. Second, because (informal) self-employment is usually regarded as an occupation of “last-resort” in developing countries (for Colombia, see Bernal 2009, 185–86).

The remainder of this article is organized as follows the next section gives an overview of the relevant self-employment and migration literature. In the third section, we present a theoretical model of self-employment. In the fourth and fifth sections, we describe the data sets and present some descriptive statistics on the main variables of interest, respectively. The econometric results are presented and discussed in the sixth section. Our conclusions are presented in the final section.

Related Literature

In the specialized literature, self-employed workers are defined as individuals who are not remunerated by a wage or a salary but who gain their income by working on their own account and bearing their own risk (Parker 2004). According to the definition, the self-employed comprise a highly heterogeneous group of workers: on one hand, there is the successful entrepreneur who runs a profitable business, invents new products, and is constantly looking for new market opportunities. On the other hand, self-employment is a survival strategy for those who are not able to find a job. In practice, most data on self-employment rely on labor force and household surveys where individuals are asked to report their employment status.

The core question in that area of research is, What motivates an individual to become self-employed? From an economist’s point of view, an individual will make a rational choice decision: He or she will prefer self-employment over the alternatives of wage employment, unemployment, or being economically inactive if the expected utility from being self-employed is greater than the utility from alternative choices, conditional on a set of attributes X:

\begin{aligned} \begin{matrix} E \{U_s e l f - e m p l o y m e n t (X)\} > E \{U_w a g e - e m p l o y m e n t (X)\} \\ E \{U_s e l f - e m p l o y m e n t (X)\} > E \{U_u n e m p l o y m e n t (X)\} \end{matrix} \\ E \{U_s e l f - e m p l o y m e n t (X)\} > E \{U_i n a c t i v e (X)\} . \end{aligned}

In industrialized countries, this model may be an adequate representation of reality: self-employed individuals could be conceived as having special “entrepreneurial skills” that, if receiving a wage offer high enough, would choose to become wage employed. However, Haile (2008) points out that this cannot be taken for granted in developing countries. Citizens of these countries, especially those with low levels of education, may only have the possibility to choose between self-employment and unemployment in many cases. As there are no unemployment benefits in the majority of low- and middle-income countries, the expected utility of being unemployed is supposed to be low enough so that individuals favor self-employment over unemployment. The choice between self-employment and being economically inactive is often relevant for the spouse and children in the household. Leibovich, Nigrinis, and Ramos (2006) observe that in Colombia secondary household members retire from labor markets when the head of household earns more.

The rational choice approach can be linked with a reduced-form model where different factors are postulated as determinants of self-employment. These can be grouped into three categories, namely, (1) monetary parameters; (2) individual abilities, tastes, and preferences; and (3) institutions and macroeconomic conditions. Perhaps, one of the most prominent arguments from the first category is the so-called earnings differential, which states that people choose to be self-employed if their expected income is higher than it would be with wage employment. Empirical evidence for this hypothesis is provided by Bernhardt (1994) for Canada, Taylor (1996) for the United Kingdom, Johansson (2000) for Finland, and Destré and Henrard (2004) for Colombia, who find evidence of negative selection into self-employment. Evidence against this hypothesis is found in Hamilton (2000), who argues that self-employed earn a lower initial income and earnings grow at a lower rate than for paid employment; hence, there must be nonpecuniary benefits as well.

Other monetary parameters that are mentioned in the literature of self-employment include initial wealth distribution (Banerjee and Newman 1993; Mesnard and Ravallion 2001; Demirguc-Kunt, Klapper, and Panos 2009) as well as access to credit and capital (Evans and Jovanovic 1989; Bernhardt 1994).

Individual abilities, tastes, and preferences are composed of attitudes to risk, education, labor market experience, family background, personal characteristics, preference for autonomy, and gender. Empirical evidence on these factors is mixed. On one hand, a number of studies for developed countries find that being white, male and married, and having labor market experience and a self-employed parent increase the probability of becoming self-employed (Hundley 2000; Eren and Sula 2009). On the other hand, González and Villarreal (2006) conclude that women in Mexico, often without substantial labor market experience, tend to favor self-employment since it can be more easily combined with household chores and looking after children than wage employment. A positive attitude to risk and preference for autonomy favors self-employment over wage employment (Hundley 2000; Hamilton 2000; Cramer et al. 2002; Fairlie 2002; Kan and Tsai 2006). Possibly, the most controversial role is the influence of education on the probability to become self-employed. Some argue that education enhances managerial ability, which increases probability of entrepreneurship while others point out that higher levels of education generate better options in wage employment reducing self-employment with rising levels of education. On the contrary, in models where informal self-employment is considered as state of last resort, it is the least educated who (involuntarily) choose this occupation (see Jacobs [2007] for a theoretical model and van der Sluis, van Praag, and Vijverberg [2005] for an overview of empirical studies for developing countries).

The impact of age is not clear-cut as age affects the probability to become self-employed through various channels. With rising age, individuals accumulate physical and human capital that makes it easier to become self-employed successfully in a challenging environment. Yet, it is also observed that older people who become unemployed and do not have a real chance to get another job in the formal labor market choose to engage in self-employment activities to earn their living.

Until now, there is just a scarce literature on the impact of conflict on labor market outcomes in general. Deininger (2003) investigates the link between civil strife and nonagricultural micro-enterprises in Uganda, concluding that violent conflict leads to a reduction in investment and the establishment of nonagricultural enterprises. Additionally, two papers deal with the effect of displacement on labor market outcomes: Kondylis (2010) compares the displaced to stayers in postwar Bosnia-Herzegovina, finding a higher unemployment rate for displaced men. Calderón and Ibáñez (2009) investigate the impact of internal refugees on labor markets at urban area destinations in Colombia. They report that wages are particularly affected in the informal sector due to an influx of additional labor. Moreover, the surge in labor supply due to the influx of displaced population in urban areas increases the likelihood of employment in the informal sector.

We also expect violent conflict not only to have an impact on wages and unemployment but also on the probability to become self-employed and/or to exit self-employment. There are various channels through which conflict, depending on the intensity, type, and consequences, may influence self-employment. The direction of the impact not only depends on the conflict but also on the economic structure of the municipality and whether the community is affected directly or indirectly by the conflict. A municipality is directly affected if it is attacked by illegally armed groups and suffers high homicide rates. As a consequence, some of the population starts leaving the municipality and relocates to other municipalities. The municipalities receiving internally displaced persons are those that are indirectly affected by the conflict. Thus, the effects of conflict are not equal across the country and there are geographical differences.

Because the civil conflict in Colombia often forces families to migrate from their place of origin (Ibáñez and Velez 2008), it is important to incorporate in our analysis previous research dealing with how inflows of migrants affects labor market conditions. While most studies cited deal with international migration, the conceptual framework can be adapted to illuminate the links between internal displacement and labor marker outcomes.

Different studies have tackled this question using US Census data (Altonji and Card 1991; LaLonde and Topel 1991; Borjas, Freeman, and Katz 1996). In his literature survey article, Borjas (1994) documents that different estimates of the elasticity of native wages with respect to the size of immigrants’ pool is negative in most cases, but these estimates are rather disperse. Others have followed “natural experiments” to investigate this impact. A prominent example is Card’s (1990) “Mariel boatlift” study, using the unexpected arrival of 125,000 Cuban immigrants to Miami to identify the impact on employment and wages of natives. The impact on the Miami labor market was barely noticeable if at all, even its impact on the year of arrival.

Another point that is relevant for us is that of skills. Altonji and Card (1991) presented an analytical framework and disaggregated the inflow of migrants into skilled and unskilled workers. This distinction is important, because the migrant inflow may affect some native workers more than others. For example, an inflow of low-skill workers will particularly affect natives who are employed in low-skill jobs. In fact, the work of Borjas, Grogger, and Hanson (2010) suggests that this effect could go beyond the labor market, eventually affecting other behavioral decisions (e.g., criminal activity) of natives with similar skills to those immigrating.

Another important point is that of “assimilation.” In his pioneering work, Chiswick (1978) used US census data to analyze how skills of migrants (and its reward in the labor market) increase with years since migration. Chiswick finds that immigrants initially receive wages that are lower than do natives of similar characteristics, but he finds that this gap closes over time. He interpreted these findings as suggestive that immigrants are “more able and highly motivated” than natives (Chiswick 1978, 900).² Although his cross-sectional results have been challenged, his study brought up the notion of “selection” of immigrants (see also Borjas [1994] for a discussion).

There are important concepts to draw from studies on internal migration as well. Using longitudinal data from the United States (NLSY), Borjas, Bronars, and Trejo (1992) found that a “wage penalty” for newly arrived internal migrants disappears in a few years. Unfortunately, the time span of our data does not allow us to observe adjustment processes that may occur. Also, one must keep in mind, though, that internal migrants in the United States are not entirely comparable to forced migrants in Colombia.

A Theoretical Model

In order to derive testable hypotheses, we consider a simple two-sector labor market model: one for wage employed individuals and another for self-employed individuals.³ Given the nature of informality in Colombia, the wage employment market is itself composed of two sectors: a formal sector, with wages w above a certain minimum wage $\underline{w}$ and the informal market, where this constraint is not binding.

In the formal wage employment market, the aggregate labor demand for an individual is defined by $w^{f} (N_{f})$ for w ≥ $\underline{w}$ or, alternatively, for $N_{f} \leq {\overset{ˉ}{N}}_{f}$ where ${\overset{ˉ}{N}}_{f}$ is such that w^f $\underline{w}$ , so that the formal wage employment market cannot absorb more workers.⁴ In the informal wage employment market, the aggregate labor demand has no quantity/wage constraints, $w^{i} (N_{i})$ . In what follows, we concentrate on the (horizontal) sum of both demands, and define the wage employment market labor demand schedule as w (N), with N = N_f + N_i . Notice that even assuming that $w^{f}$ and wⁱ have continuous derivatives defined in their respective domains, the wage employment could have a “kink” (discontinuity in the derivative) at N such that $w (N) = \underline{w}$ .

In the self-employment market, the Nth individual entering such market will obtain an earning defined by e(N). Labor demand for this and the wage employment markets can also depend on other factors (e.g., local demand conditions, the price of other factors, individual characteristics such as skills). We omit them here for simplicity.

The consequences of an influx of potential workers have different effects on the labor market. First, the supply of newly arrived displaced individual (potential workers) increases the total labor supply. Second, this influx can also shift labor demand. This is because the “village” demand for final goods increases and thus it affects input (e.g., labor) demand. The shift in the demand for goods curve may be small, especially since migrants may be asset and wealth deprived (at least immediately after displacement).⁵

At this point, it is important to introduce skills in the model; that is, we allow for individual heterogeneity. Following the literature on migration, it is likely that the impact of the arrival of workers will affect sectors of the labor market that more closely match displaced individuals’ skills. For example, the arrival of unskilled workers may particularly affect those residents working in low-skill jobs and may not affect those natives with stronger skill profiles. In fact, informal and self-employment jobs in the rural labor market can be characterized as low-skilled jobs (Leibovich, Nigrinis, and Ramos 2006). We can check this in our data as well. For this, one would like to compare conflict-displaced with natives. We construct two definitions of “conflict-displaced” and “native” (or stayers).⁶ Using t-tests, we observe that for both classifications, individuals ages fifteen to sixty who are conflict-displaced are—on average—less educated than and of the same age as natives. This suggests that conflict displaced in our sample have an adverse skills profile compared with natives in our sample. In summary, if migrants are relatively unskilled, we expect them to look for jobs in the informal or self-employment sector, driving wages down in those sectors.

In the model described earlier, the impact of an influx on migrants on the relative size of the self-employment sector is ambiguous. Migrants may allocate themselves to the informal (or even formal) wage employment sector. So far, the direction and magnitude of this impact depends on the slope of the e(N) and w(N). However, one must bear in mind that the model displayed earlier simply assumes that the migrant’s decision to “select into” wage or self-employment sectors is unconstrained. Upon arrival, according to the simple model, migrants are supposed to compare prospects in these sectors and decide. This may not be the case. It may not be easy for newcomers to have “connections” to immediately receive wage offers. In fact, using data from the Encuesta Continua de Hogares (ECH), Bernal (2009, 186) indicates that “the most prevalent reason, for both men and women, to work as self-employed is that they could not find another job.” If this is the case, one may expect the share of self-employment to increase in areas receiving migrants (at least in the short run). We can test this using our data set.

Another impact that we are interested in is that of attacks on labor market outcomes. What effect does an attack have on labor market outcomes? Labor demand, being an input demand depends on economic conditions. Different studies show a correlation between economic prosperity (growth, firm’s sale) and markers of conflict (political instability or criminality), but these study topics are not directly linked with Colombia’s civil war either because they are estimated with a cross section of countries or because they rely on sample evidence from other countries (e.g., Alesina and Perotti 1996; Gaviria 2002; Abadie and Gardeazabal 2003).

The two studies that are most informative for understanding the link between conflict and economic activity in Colombia are those by Pshisva and Suarez (2010) suggesting a link between kidnaps and firm investment and Camacho and Rodriguez (2010), which use two Colombian longitudinal data sets to find that attacks at the municipality level increase the probability of firm exit. Their estimates also take into account the potential endogeneity of conflict. In sum, it is reasonable to argue that labor demand schedules may be shifted inward when attacks occur. The return to self-employment e(N) may also be affected in the same fashion. It is not clear, a priori what to expect. In particular, there may be out-migration from the attacked area, affecting the supply of labor. But if we assume that labor supply remains constant, and with self-employment as a last-resource option, it is possible to hypothesize that conflict may increase the share of self-employment and decrease wages. By affecting law and order, conflict may shift the composition of the labor market and tilt it toward more informal contractual arrangements, among them self-employment, where the entrepreneur bears all the risk.

The Data

In this study, we use three types of data: (1) a household survey by the Familias en Acción program, (2) a municipality-level data set on violence and conflict, and (3) a data set describing the economic situation of municipalities. The first data set was established in order to analyze the effects of a Conditional Cash Transfer (CCT) program on nutrition, health, and education of poor children aged 0 to 17 implemented by the Colombian government, the World Bank, and the Inter-American Development Bank. The baseline survey was conducted in 2002, the first follow-up carried out in 2003 and the second follow-up in 2005 or 2006. We used the first and the sixth module of the survey for our analysis. In these modules, information about the socioeconomic structure of the household, housing conditions, household assets, education, access to infrastructure, usage of health care services, household consumption, labor supply, income, and transfers were collected. Although the information is extensive and is of good quality, one must bear in mind that the sample is not representative of the overall rural population.

The second data set, assembled by the Center of Economic Development Studies (CEDE) at the Universidad de los Andes (Bogotá, Colombia), includes information about violence and conflict intensity and contains municipality characteristics and time-municipality variant instrumental variables (all of which are discussed in more detail later). These characteristics include the department the municipality is located in, the total number of inhabitants of each municipality, and the share of urban and rural population at municipality level. Since the homicide rates are missing for the years 2005 and 2006, we augment this data set with data on homicide rates obtained from the National Administrative Department of Statistics (DANE) and the National Police.

The third data set comes from Colombia’s National Planning Department (DNP) and comprises information on the municipality’s industrial and commercial taxes (ICA). Since taxes are reported in nominal Colombian pesos, we converted them into real Colombian pesos using the Consumer Price Index (CPI) calculated by DANE. Tax collection indicators capture the municipality’s economic situation, which affects labor demand and may impact the level of violence.

Descriptive Statistics: Displacement and Labor Market Transitions⁷

To better understand dynamics, we study labor market transitions in two different groups of adults. One is a set of individuals considered “Not displaced,” with no mobility across waves and living in a receiving district.⁸

Panel A in Table 1 shows a transition (or stochastic) matrix. The first column displays the labor market status of the person in the first wave, whereas the first row displays the status of the person in the last (third) wave. The four labor market statuses define a 4 × 4 matrix. Let A_i _,j be the element in the ith row and jth column of that matrix. These cells display conditional probabilities p_i _,j = prob (Status in third wave = j|Status in first wave = i). Indexes j and i represent the four possible labor market statuses. The last row display the unconditional probability of belonging to a given labor market status in wave 3.

Table 1.

Displacement and Employment Status.

First wave	Unemployed	Self-mployed	Employed	Inactive	Total
Panel A Sample: Not displaced (number of observations: 12,329)
Unemployed	13.39	18.64	44.62	23.36	100
Self-employed	2.69	48.77	31.08	17.46	100
Employed	3.37	18.01	59.07	19.54	100
Inactive	2.49	9	19.14	69.37	100
Total	3.2	23.8	37.71	35.29	100
Panel B Sample: Displaced (Number of observations: 1,739)
Unemployed	4.88	29.27	46.34	19.51	100
Self-employed	2.29	48.17	34.86	14.68	100
Employed	2.51	19.08	61.83	16.57	100
Inactive	1.54	9.9	18.09	70.48	100
Total	2.19	23.52	39.97	34.33	100
Panel C Sample: Stayers in expulsing communities (number of observations 18,974)
Unemployed	10.27	24.33	43.75	21.65	100
Self-employed	1.76	51.67	30.58	15.99	100
Employed	2.95	21.54	58.69	16.82	100
Inactive	1.93	12.71	17.64	67.72	100
Total	2.46	27.25	37.17	33.12	100

Note: The survey does not indicate the displacement status of individuals. However, we used migrationtrajectories to classify individuals. In particular, individuals that moved in/out from cabecera or within municipalities between waves 1 and 3, and that at baseline lived in expulsing (net displacement < 0) municipalities, are classified as “displaced.” Individuals whose cabecera (living in the administrative city of municipality) status and municipality remained unchanged throughout the survey and who lived in receiving municipalities (net displacement > 0) are deemed “not displaced”. Finally, individuals whose cabecera (living in the administrative city of municipality) status and municipality remained unchanged throughout the survey and who lived in expulsing municipalities (net displacement > 0) are deemed “stayers in expulsing communities.” The transition matrices uses information for individuals ages fifteen to sixty.

Equipped with these definitions, we observe that the labor market segmentation (in these categories) is similar in both samples; that is, the proportion of unemployed, self-employed, employed and inactive individuals is similar in wave 3. However, the transitions could be different.

There are many comments to make, but focusing on the most important remarks, one can observe that the transition from unemployment to self-employment is higher in displaced (29 percent, panel B) than in not displaced (19 percent, panel A) individuals. This suggests that their mobility was linked to a change in labor market status, being more likely (than not-displaced persons) to become self-employed. Notice also the persistence of employment and self-employment. About 50 to 60 percent of these individuals remain in the same category that they were in wave 1 (baseline). We also focused (results not shown) on not-displaced persons living in municipalities that at baseline were receiving displaced individuals to a large extent (net displacement > 1,000 per 100,000 inhabitants). Although the sample size is reduced from 12,329 observations to 2,901, we do not find substantial differences with respect to panel A. This suggests that the influx of individuals (to the magnitude observed in our case) does not seem to largely affect the labor status of locals.

Individuals in panel C only differ from those in panel A in that they were living in expulsing (not receiving) communities. So it is informative to compare panel C with panel A. Looking at conditional probabilities, and for any state in wave 1, stayers in expulsing communities are more likely to transit into self-employment and less likely to transit into inactivity. However, the employment retention rate (transit from employed to employed) in panel C is as high as in panel A. Of course, this is confounded by the effect that individuals who lost their wage employment may have moved elsewhere, so this and others are, to a certain extent, samples driven by labor market status.

Estimation Strategy

In this section, we investigate the impact of conflict on self-employment econometrically. Our theoretical framework indicated that labor market outcomes (wages, employment) depended on migration inflows and variables describing the conflict environment. Moreover, other covariates that affect outcomes can be included, for example, market conditions in the municipality, and individual and household characteristics.

As a first approximation, one could use fixed effects (OLS-FE) estimation to exploit the panel data structure to control for time-invariant individual heterogeneity, which may bias cross-sectional results. We run regressions of the form

y_{i j t} = β_{0} + X_{i j t} β + C_{j t} γ + α_{i} + β_{t} + u_{i j t},

where y_ijt is an indicator of self-employment (or another outcome of interest) for individual i living in municipality j at time t, X_ijt is a vector of individual, household, and municipality controls, C_jt includes our vector of conflict variables net displacement (defined as difference of receiving and expulsing people in a municipality) and attacks, α _i captures a time-invariant unobserved individual effect, β _t captures systematic variation across time (time-fixed effect), and u_ijt is the usual error term, which is allowed to be heteroscedastic in our estimations. The sample is restricted to individuals aged ten years or above. Our parameter of interest is captured in the vector γ, giving an estimate of the effect of net displacement or attacks on the probability of being self-employed.⁹

There is evidence that the occurrence of conflict may be correlated with the error (e.g., for the case of Colombia, see Rodríguez and Sánchez [2009]; and Camacho and Rodriguez [2010] among others). Under this hypothesis, fixed-effects estimates will be inconsistent.

One possibility is using lagged values of measures of conflict, which are predetermined variables. However, this does not per se imply that this new approach will yield consistent estimates since serial correlation may be present, making lags correlated with the contemporaneous error u_ijt .

A second strategy to overcome the potential inconsistency of estimators is to use instrumental variables (IV-FE). Because both net displacement and attacks are potentially endogenous, this constrains us to use at least two instruments. Throughout our estimations, we use three instruments that vary across time and municipality levels. These instruments have been used in previous studies about conflict in Colombia and are arguably suitable for our case. First, we use an indicator of central government deterrence measures: (lag) the rate of homicide captures¹⁰ at the state level, interacted with the respective municipal population. This is the approach followed by Rodríguez and Sánchez (2009). This instrument is valid if homicide captures are not directly correlated with labor market outcomes. This is plausible since, as Rodríguez and Sánchez (2009) argue, deterrence decisions are under “central government control (the Ministry of Defense).” In addition to this, since the deterrence factor is at the state level (and thus not easily observed by households), it is difficult to come up with reasons why this would affect household behavior in the labor market. Second, we follow Camacho and Rodriguez (2010) and use (lags) of laboratory dismantle and antinarcotic operations. These two instruments act as a measure of “presence and effectiveness of the government to counteract criminal activity” (Camacho and Rodriguez 2010, 12).

In sum, our estimation strategy will compare three different estimates, one from fixed-effects panel data (OLS-FE, where we control for individual fixed effects, time fixed effects, and departmental fixed effects), estimates using lagged conflict indicators instead of contemporaneous regressors, and instrumental variables (IV-FE) estimates.

IV estimates result from instrumenting conflict variables (attacks, net displacement per 100,000 inhabitants) in a first-stage regression of the form

C_{j t} = β_{0} + Z_{j t} β + α_{j} + β_{t} + ϵ_{j t},

where C is the suspected endogenous conflict indicator (attacks, net displacement) Z are exogenous regressors (from 1 and including the instruments).

Table 2 shows the first stage estimates for equation (2) with the two potential endogenous variables and showing only the coefficient for the three instruments.¹¹ The homicide capture rate proxies an exogenous “deterrence” effect (Rodríguez and Sánchez 2009). Surprisingly, homicide capture rates have the opposite sign for net displacement. In other words, a high capture rate in the previous year predicts lower displacement inflows this year. First, one must notice that the coefficient is very small: going from quantile 25 to 75 of capture rates in the sample would predict a drop in the net displacement of 72 individuals per 100,000, which is small, compared to the standard deviation of net displacement, which is about 1,359 individuals/100,000. Also, it is possible that such deterrence could “expulse” from the community individuals involved in illegal activities.¹² Regarding the effect of homicide captures on attacks, its effect is negative, but not statistically significant. The lag of antinarcotic operations increases net displacement, but its effect is not significant. However, it has a negative effect on attacks, which is intuitive following the argument that it is an indicator of deterrence.

Table 2.

First-stage Regressions.

Dependent variable	Net displacement	Attacks
Laboratories Dismantle (t − 1)	131.4799***	0.1055*
	(11.6760)	(0.0588)
Antinarcotics Operations (t − 1)	3.2134	−0.8952***
	(7.0016)	(0.1160)
Homicide Captures (t − 1)	−0.0088***	0.0000
	(0.0019)	(0.0000)
F-test	51.92	23.47
Observations	26,893	26,893
R ²	0.1022	0.1985
Number of groups	10,816	10,816

Note. Estimation with department and time dummies and with robust standard errors (below estimates).

***p < .01. **p < .05. *p < .1.

The lag of lab dismantling operations has a positive effect on the inflow of displaced individuals: the “crime deterrent” effect may render the place “safe” for individuals displaced from conflict. However, we find that lab dismantling increases the number of attacks. It is possible that the presence of armed groups trigger a reaction from individuals in illicit activities, and thus, it is possible to expect also a positive relationship (Camacho and Rodriguez 2010).

In sum, our first stage estimates appear to be relevant overall, although we test that formally later, using appropriate statistics, given that we have two (potentially) endogenous regressors.¹³

Now we compare the results of estimates showing the impact of conflict on self-employment presented in Table 3. We present the results in three groups of regressions. Columns 1 to 3 present the results of fixed effect regressions, using contemporaneous indicators of conflict (attacks and net displacement). Specification 1 has both conflict indicators and the contemporaneous homicide rate (to control for violent crime). The second column excludes homicide rates, while the third includes homicide rates but drops attacks. These three columns allow us to gauge the stability of estimates and to select a “preferred specification.” The theoretical model was not clear-cut regarding predictions, but displaced individuals (which we identified as “movers”) had a poor skill background (both in our sample and as suggested in the literature), which may make them less likely to be offered wage work in the short run. This may also be valid if one assumes “lack of connections” to obtain paid jobs immediately upon arrival. Our theoretical model was neither clear regarding the impact of attacks on self-employment. Under certain assumptions (if wage work demand shifts inward and/or labor supply remains constant), one may find an increment of the share of self-employment to occur in response to attacks.

Table 3.

Results of Violent Conflict on Self-employment.

Estimation strategy	OLS-FE	OLS-FE	OLS-FE	OLS-FE	OLS-FE	OLS-FE	OLS-IV	OLS-IV	OLS-IV
Dependent variable	Self-employment
	1	2	3	4	5	6	7	8	9
Net displacement	0.00001***	0.00001***	0.00001***				0.00005**	0.00005**	0.00006***
Net displacement	(0.00000)	(0.00000)	(0.00000)				(0.00002)	(0.00002)	(0.00002)
Attacks	0.00074	0.00042					0.01580***	0.01540***
Attacks	(0.00078)	(0.00074)					(0.0057)	(0.00575)
Net displacement (t − 1)				0.00000**	0.00000**	0.00000**
Net displacement (t − 1)				(0.00000)	(0.00000)	(0.00000)
Attacks (t − 1)				−0.00289***	−0.00300***
Attacks (t − 1)				(0.00092)	(0.00089)
Homicide rate	−0.0001		−0.00007	−0.00005			−0.00050**		0.0001
	(0.00008)		(0.00008)	(0.00008)			(0.00023)		(−0.00011)
Homicide rate (t − 1)						−0.00013
Homicide rate (t − 1)						(0.00008)
Age	−0.00253	−0.00243	−0.00263	−0.0029	−0.00281	−0.00273	0.00094	0.00144	−0.00128
Age	(0.00595)	(0.00595)	(0.00595)	(0.00594)	(0.00594)	(0.00595)	(0.00621)	(0.00622)	(−0.00605)
Age²	0.00001	0.00001	0.00001	0.00002	0.00001	0.00001	−0.00002	−0.00002	0
Age²	(0.00006)	(0.00006)	(0.00006)	(0.00006)	(0.00006)	(0.00006)	(0.00007)	(0.00007)	(−0.00006)
Treatment	0.13200***	0.13200***	0.13200***	0.13600***	0.13600***	0.13500***	0.12800***	0.12900***	0.13600***
Treatment	(0.0124)	(0.0124)	(0.0124)	(0.0124)	(0.0124)	(0.0124)	(0.0134)	(0.0133)	(−0.0128)
Ill in last 2 weeks	0.01940**	0.01950**	0.01930**	0.01910**	0.01920**	0.01910**	0.02060**	0.02130**	0.01890**
Ill in last 2 weeks	(0.00897)	(0.00897)	(0.00897)	(0.00898)	(0.00898)	(0.00898)	(0.00917)	(0.0092)	(−0.00902)
Dependency ratio	0.00856	0.00855	0.00854	0.00773	0.00771	0.00783	0.0105	0.0102	0.0101
Dependency ratio	(0.00684)	(0.00684)	(0.00684)	(0.00686)	(0.00686)	(0.00686)	(0.00704)	(0.00703)	(−0.00686)
Log (household size)	−0.06470***	−0.06450***	−0.06400***	−0.06390***	−0.06400***	−0.06630***	−0.06690***	−0.06870***	−0.05310**
Log (household size)	(0.0236)	(0.0235)	(0.0235)	(0.0235)	(0.0235)	(0.0235)	(0.025)	(0.0251)	(−0.024)
Cabecera	0.0126	0.0123	0.0122	0.0128	0.0128	0.0128	0.0192	0.0185	0.00907
Cabecera	(0.0197)	(0.0197)	(0.0197)	(0.0197)	(0.0197)	(0.0197)	(0.0207)	(0.0207)	(−0.0199)
Log (Population)	−0.05670**	−0.05720**	−0.05260*	−0.0287	−0.031	−0.05010*	−0.10800**	−0.10700**	−0.0193
Log (Population)	(0.0286)	(0.0284)	(0.0283)	(0.0301)	(0.0299)	(0.0282)	(0.0461)	(0.0449)	(−0.0309)
Log (ICA/capita)	0.01360**	0.01300**	0.01340**	0.01250**	0.01200**	0.01320**	0.01730***	0.01430**	0.01340**
Log (ICA/capita)	(0.00574)	(0.00573)	(0.00574)	(0.00576)	(0.00575)	(0.00574)	(0.00594)	(0.00574)	(−0.00572)
Constant	1.23300***	1.22900***	1.19400***	0.73400**	0.98000***	0.97000***
Constant	(0.321)	(0.319)	(0.318)	(0.33)	(0.331)	(0.31)
Observations	26,893	26,894	26,893	26,893	26,894	26,893	26,893	26,894	26,893
Number of groups	10,816	10,816	10,816	10,816	10,816	10,816	10,816	10,816	10,816
Weak identification test
Kleibergen-Paap rk Wald F statistic							22.399	20.232	51.921
Overidentification test df							1	1	2
Hansen J statistic							1.854	2.056	13.165
p value							.1734	.1516	.0014
Endogeneity test df							2	2	1
χ²							19.012	18.489	9.756
p value							.0001	.0001	.0018

Note: The table reports estimates for the probability to be self-employed conditioning on being employed. We use a binary dependent variable, taking the value 1 if a person is self-employed and 0 for wage employment. All individuals aged ten and older belonging to the working population were included in the sample. Cabecera is a dummy variable taking the value 1 if municipality’s cabecera to account for urban–rural differences. Treatment is a dummy for participation in Familias en Acción. Individual, department, and time-fixed effects are included. Robust standard errors in parentheses.

***p < .01. **p < .05. *p < .1.

Looking at columns 1 through 3, we observe that only net displacement is significant, with a positive sign. To interpret the magnitude of the effect, we proceed as follows. Given the estimates (which are similar across these specifications), a one standard deviation increase in net displacement (1,359 individuals per 100,000 inhabitants in our regression sample) will increase the rate of self-employment by about 1.5 percent points. The mean rate of self-employment in the sample used in this regression is about 41 percent. However, one should bear in mind that a one standard deviation increase in net displacement would cause an influx of about 1.3 percent points (1,359/100,000) in the population of a given municipality. So a 1.5 percent point increase in the rate of self-employment/total employment is a large magnitude taking into account that not all displaced individuals are in the active age group or would be looking for work.

Columns 4 through 6 follow the same pattern as columns 1 through 3, but this time using lagged conflict variables (and homicide rates). Here, the effect of lagged net displacement is smaller in magnitude but preserves the sign and the statistical significance. Attacks (lagged) reduce self-employment. To gauge the relevance of the coefficient, we calculate the (sample based) interquartile range of attacks, which is four units, so that four additional attacks reduce (given the estimates) the rate of self-employment by 1.1 percent points. However, both fixed-effect strategies (with contemporaneous or lagged conflict variables) yield inconsistent estimates in case of endogeneity. In the case of lagged regressors, estimates will be inconsistent given serial correlation in errors, a case that is often present in panel data.

Therefore, in columns 7 through 9, we present IV estimates. The presentation of estimates across columns follows the same patterns as in columns 1 through 3 and 4 through 6. The impact of net displacement on self-employment rates (ratio of self-employed to all employed individuals) is now about five times as larger as in the columns 1 through 3. In our case, with two endogenous variables (and several controls and fixed effect), one cannot easily sign the direction and magnitude of the biases to be expected.¹⁴ In one endogenous regressor cross-sectional OLS case, this (negative) bias would arise if the endogenous regressor is negatively correlated with the error (in a probability limit sense). It is also possible that if measurement error is present, the “attenuation” bias is also corrected by the IV-FE strategy, so one would expect estimates to “move away” from zero in IV versus OLS. Notice that this comes at a price: standard errors are substantially larger. Column 7 includes Homicide rates as control, which is not significant. Column 9 excludes Attacks, which is significant. Therefore, 8 is our preferred IV-FE specification. Using these estimates, a one standard deviation increase in net displacement rates (1,359 migrants per 100,000) would increase the rate of self-employment by about 7 percent points, a large fraction given that, as we said, the rate of self-employment is 41 percent in this regression sample. The coefficient of attacks is of the same sign than that of net displacement. As before, using the interquartile range of attacks, four additional attacks would increase the rate of self-employment by 6.2 percent pints (four times 0.01540).

It is important to be cautious about these findings because IV-FE estimates rely on specification assumptions. This calls for testing the relevance of instruments (using a weak-instrument test) and a test for “exogeneity” of instruments. In terms of instrument relevance, the recommended test, given that we have two endogenous regressors, is to use a Kleibergen-Paap Wald-type statistic, which is a generalization of the Cragg-Donald statistic. It is well known that IV estimates are biased and liable to size distortions. However, the larger the Kleibergen-Paap statistic, the smaller the bias/size distortions will be. The values are often compared against tables in Stock and Yogo (2005). In our case, accepting a (minimal) size distortion of 10 percent would require a Kleibergen-Paap above 13.43. In our case, the statistic is 20.23, so our instruments can be considered “strong” enough so that IV bias and size distortions are minimal. The overidentification test is informative in cases where we have more excluded regressors than endogenous variables. Although it is a diagnostic test of “exogeneity” of instruments, it does not test the whole set of them, but just assumes that there are K exogenous instruments (K = number of endogenous variables) and test whether the remaining ones are exogenous. The Hansen J-Statistic is appropriate for our case. Because we have three instruments and two endogenous variables in equation (8), the J-stat is distributed as χ²(1). We do not reject the null that the additional instrument is exogenous, so on this ground, the exclusion restriction appears valid.¹⁵ Finally, one can check the null of exogeneity of the variables suspected of being endogenous. This is performed using an exogeneity test, which rejects the null in our case, as shown at the bottom part of column 8.

In sum, we have presented IV estimates that suggest that indeed conflict variables are endogenous. Moreover, we check that our instruments are relevant and valid. The estimates are larger than our fixed-effect estimates (assuming exogeneity). The magnitude of the coefficients is large and indicates that displacement and conflict have a strong effect in our context of study.

Next, we investigate if the effects of conflict on entrepreneurship differ by type of activity. For this purpose, we created two indicators. First, we created an indicator that is 1 if the individual is self-employed in services and 0 if the individual is employed (or self-employed) in other activity.¹⁶ For comparison, in column 1 of Table 4, we display OLS-FE estimates. IV-FE estimates in column 2 through 4 are to be preferred because of evidence of endogeneity, instrument relevance, and validity (as explained above for self-employment regressions). It appears that for services, net displacement increases the fraction of individuals in self-employment; 14 percent of the individuals in the regression sample are self-employed in the service sector. Using an average estimate from specifications 2 to 4, a 1 percent standard deviation in net displacement explains an increase of 5 percent points in this ratio, a large effect. This is plausible if a large influx of migrants reaches the cabecera (or the administrative town of the municipality, where such types of self-employment are usually located)¹⁷ and stays there. If their opportunities in wage employment are initially scarce (due to lack of skills, connections), a large part of them may choose to work in the self-employment sector. No significant effect is found for attacks.¹⁸ Second, we created an indicator that is 1 if the individual is self-employed in agriculture and 0 if the individual is employed (or self-employed) in other activity. As before, column 5 portrays OLS-FE estimates. Columns 6 to 8 present IV-FE estimates. No effect is found for net displacement. However, attacks, which may be more frequent outside the cabecera, could—in the short run—leave individuals with the only option of self-employment. This requires low mobility, which is the case in our sample, unless conflict escalates to a high level of violence (see Mesnard 2009), and a reduction in the demand for labor in wage employment (creating a “surplus” of workers allocated to the self-employment sector).

Table 4.

Results of Violent Conflict on Self-employment.

Estimation strategy	OLS-FE	OLS-IV	OLS-IV	OLS-IV	OLS-FE	OLS-IV	OLS-IV	OLS-IV
Dependent variable	Self-employed in services				Self-employed in agriculture
	1	2	3	4	5	6	7	8
Net displacement	0.00000	0.00004***	0.00004***	0.00004***	0.00001***	0.00001	0.00001	0.00002
Net displacement	(0.00000)	(0.00001)	(0.00001)	(0.00001)	(0.00000)	(0.00002)	(0.00002)	(0.00002)
Attacks	0.00224***	0.00332	0.00327		−0.00155**	0.01830***	0.01800***
Attacks	(0.00053)	(0.00214)	(0.00214)		(0.00062)	(0.00631)	(0.00641)
Homicide rate	−0.00017***	−0.00007		0.00006	0.00004	−0.00071***		0.00001
Homicide rate	(0.00006)	(0.00011)		(0.00008)	(0.00007)	(0.00026)		(0.00010)
Age	0.00147	0.00271	0.00277	0.00217	0.00581	0.00922	0.00991*	0.00624
Age	(0.00367)	(0.00375)	(0.00376)	(0.00373)	(0.00535)	(0.00582)	(0.00583)	(0.00549)
Age²	−0.00001	−0.00003	−0.00003	−0.00002	−0.00011*	−0.00014**	−0.00015**	−0.00012*
Age²	(0.00004)	(0.00004)	(0.00004)	(0.00004)	(0.00006)	(0.00006)	(0.00006)	(0.00006)
Treatment	0.03330***	0.03620***	0.03650***	0.03820***	0.10800***	0.09690***	0.09980***	0.10700***
Treatment	(0.00832)	(0.00860)	(0.00857)	(0.00865)	(0.01110)	(0.01290)	(0.01260)	(0.01160)
Ill in last 2 weeks	0.00252	0.00189	0.00198	0.00164	0.01470*	0.01600*	0.01710**	0.01470*
Ill in last 2 weeks	(0.00659)	(0.00661)	(0.00662)	(0.00661)	(0.00820)	(0.00854)	(0.00859)	(0.00827)
Dependency ratio	0.00366	0.00510	0.00500	0.00520	0.00684	0.00646	0.00554	0.00705
Dependency ratio	(0.00506)	(0.00518)	(0.00519)	(0.00516)	(0.00594)	(0.00640)	(0.00643)	(0.00603)
Log (household size)	−0.0350**	−0.02780	−0.02810	−0.02560	−0.04270**	−0.05480**	−0.05790**	−0.04240**
Log (household size)	(0.01730)	(0.01790)	(0.01800)	(0.01770)	(0.02100)	(0.02280)	(0.02300)	(0.02130)
Cabecera	0.00758	0.00616	0.00596	0.00422	−0.00836	0.00285	0.00129	−0.00780
Cabecera	(0.01320)	(0.01340)	(0.01340)	(0.01340)	(0.01680)	(0.01800)	(0.01790)	(0.01680)
Log (Population)	−0.07420***	−0.05770**	−0.05620**	−0.03970*	0.00591	−0.09790**	−0.09860**	0.00130
Log (Population)	(0.02080)	(0.02400)	(0.02350)	(0.02160)	(0.02090)	(0.04710)	(0.04600)	(0.02400)
Log (ICA/capita)	0.01250***	0.01290***	0.01250***	0.01190***	0.00049	0.00658	0.00241	0.00098
Log (ICA/capita)	(0.00417)	(0.00437)	(0.00434)	(0.00431)	(0.00518)	(0.00575)	(0.00539)	(0.00525)
Constant	0.69900***				0.33100
Constant	(0.23000)				(0.25600)
Observations	22,606	22,606	22,607	22,606	22,606	22,606	22,607	22,606
Number of groups	9,340	9,340	9,340	9,340	9,340	9,340	9,340	9,340

Note: The table reports estimates for the probability to be self-employed in the agricultural or services sector conditioning on being employed. We use a binary dependent variable, taking the value 1 if a person is self-employed in agriculture or services, respectively, and 0 for wage employment and self-employment in other sectors. All individuals aged ten and older belonging to the working population were included in the sample. Cabecera is a dummy variable taking the value 1 if municipality’s cabecera to account for urban-rural differences. Treatment is a dummy for participation in Familias en Acción. Department and time dummies included. Robust standard errors in parentheses.

***p < .01. **p < .05. *p < .1.

Finally, we use income and time allocation modules to calculate log(wage) in the wage and self-employment sectors. Table 5 presents the estimates of OLS-FE and IV-FE estimations. As before, column 1 presents OLS-FE estimates, whereas columns 2 through 4 present IV-FE estimates. Although instruments appear relevant and valid in the diagnostic tests, there is no evidence of endogeneity of net displacement and attacks in these equations. In other words, OLS-FE and IV-FE estimates are not substantially different. In contrast to these results, we find a substantial (negative) effect of the influx of displaced individuals in the log hourly income of self-employed individuals. The estimates displayed in columns 7 through 9 are stable and in all cases statistically significant. An influx of displaced of 1 standard deviation (1,359 displaced/100,000 inhabitants) decreases hourly incomes in the self-employment sector by about 0.7 log units, which implies that hourly incomes proceeding from self-employment are halved in this case.¹⁹ This table shows an interesting contrast between conflict-related income shocks in both sectors (self-employed vs. wage-employed). This indicates that the inflow of migrants affects skilled and unskilled workers differently. Although wage and self-employed individuals are a heterogeneous group, it is very likely that the skills of self-employed individuals are—on average—lower than those wage employed, as suggested previously.

Table 5.

Results of Violent Conflict on Hourly Income.

Estimation strategy	OLS-FE	OLS-IV	OLS-IV	OLS-IV	OLS-FE	OLS-IV	OLS-IV	OLS-IV
Dependent variable	Log (income/hour of employed)				Log (income/hour of self-employed)
	1	2	3	4	5	6	7	8
Net displacement	0.00000	0.00017	0.00015	0.00018	0.00000	−0.000502**	−0.000521**	−0.000520**
Net displacement	(0.00001)	(0.000145)	(0.00013)	(0.00016)	(0.00002)	(0.00021)	(0.00021)	(0.00020)
Attacks	−0.00144	−0.00553	−0.00440		−0.00313	−0.0252	−0.019
Attacks	(0.00183)	(0.00805)	(0.00724)		(0.00379)	(0.02940)	(0.02560)
Homicide rate	0.00025	0.00087		0.00066	0.000764**	0.00139		0.00066
Homicide rate	(0.00024)	(0.00071)		(0.00049)	(0.00030)	(0.00100)		(0.00055)
Age	0.03440**	0.03620**	0.03560**	0.03610**	0.03590	0.00757	0.00846	0.00820
Age	(0.01500)	(0.01550)	(0.01550)	(0.01550)	(0.03160)	(0.03540)	(0.03570)	(0.03570)
Age²	−0.00056***	−0.00058***	−0.00058***	−0.00058***	−0.00029	−0.00003	−0.00003	−0.00003
Age²	(0.00018)	(0.00019)	(0.00019)	(0.00019)	(0.00031)	(0.00034)	(0.00034)	(0.00034)
Treatment	−0.02390	0.03280	0.03640	0.03550	0.396***	0.321***	0.311***	0.312***
Treatment	(0.03050)	(0.05860)	(0.06040)	(0.06150)	(0.08870)	(0.09420)	(0.09350)	(0.09360)
Dependency ratio	−0.00820	−0.00421	−0.00369	−0.00253	0.0611***	0.0526*	0.0529*	0.0574**
Dependency ratio	(0.01670)	(0.01790)	(0.01780)	(0.01850)	(0.02330)	(0.02820)	(0.02830)	(0.02760)
Log (household size)	0.01790	0.05520	0.05580	0.05270	−0.07110	−0.15600	−0.17600	−0.19000
Log (household size)	(0.05360)	(0.06690)	(0.06670)	(0.06570)	(0.09940)	(0.13300)	(0.12800)	(0.11800)
Cabecera	0.09780**	0.09990**	0.10100**	0.10200**	0.171*	0.242**	0.253**	0.260**
Cabecera	(0.04060)	(0.04090)	(0.04070)	(0.04150)	(0.09890)	(0.11900)	(0.11800)	(0.11700)
Log (Population)	0.07090	0.2770	0.25300	0.25400	0.15500	−0.38700	−0.41200	−0.457*
Log (Population)	(0.06710)	(0.2070)	(0.18900)	(0.18300)	(0.15000)	(0.30300)	(0.28900)	(0.27500)
Log (ICA/capita)	−0.00019	−0.00211	0.00275	−0.00148	0.0407*	−0.02440	−0.00923	−0.02980
Log (ICA/capita)	(0.01340)	(0.01350)	(0.01340)	(0.01350)	(0.02400)	(0.04210)	(0.04480)	(0.04060)
Observations	7,900	7,900	7,900	7,900	6,642	6,642	6,644	6,642
Number of groups	3,475	3,475	3,475	3,475	2,986	2,986	2,987	2,986

Note: The table reports estimates for log (income/hour) for employed and self-employed individuals. All individuals aged ten and older belonging to the working population were included in the sample. Cabecera is a dummy variable taking the value 1 if municipality’s cabecera to account for urban-rural differences. Treatment is a dummy for participation in Familias en Acción. Department and time dummies included. Robust standard errors in parentheses. The regression for Income self-employment also includes Region × Wave dummies.

***p < .01. **p < .05. *p < .1.

Conclusion

Do violent conflicts impact (informal) self-employed workers in developing countries? Our aim is to make a first attempt at resolving this question by analyzing the effects of conflict on the probability to be a self-employed worker in rural Colombia. We compiled a data set combining information from three preexisting data sets to obtain information on individuals, households, and municipalities on one hand and conflict information on the other hand. As indicators of conflict, we include net displacement rates and the number of attacks by illegal armed groups. Furthermore, we divide the self-employed in two categories: being self-employed in services and being self-employed in the agricultural sector. With this approach, we derive four main findings for the case of Colombia.

First, the share of self-employed individuals is higher in municipalities that experience an increase in net displacement and attacks by rebel groups or paramilitaries.

Second, net displacement increases the probability to be self-employed in the services sector, but attacks do not have a significant impact on such self-employment. One explanation for this result may be that self-employment in services mainly takes place in the cabecera (urban part of a municipality), which receive a larger influx of the displaced than do purely rural areas. Unskilled migrants arriving at the cabecera often have no other opportunity to participate in the labor market than being self-employed in the services sector. Consequently, the share of self-employed in that sector increases in net receiving communities.

Third, net displacement has no significant impact on self-employment in the agricultural sector in contrast to self-employment in services. However, attacks increase the likelihood to be self-employed in agriculture significantly. This pattern could be explained by the fact that displacement mainly occurs from rural areas to cities (and therefore has no significant impact on self-employment in agriculture) while attacks take place in the countryside. In rural regions experiencing a certain number of attacks employment opportunities other than self-employment in agriculture might be scarce, and that is why agricultural self-employment increases in rural, high-conflict regions.

Fourth, income of the self-employed decreases significantly in municipalities which receive a large number of internally displaced people. As pointed out earlier on, these migrants are, on average, unskilled and often their only possibility to enter the labor market is via self-employment. Self-employment in Colombia usually takes place in the informal sector where wages and income are flexible. Thus, a large inflow of people leads to a decrease in income in that sector, which we also observe in our sample.

These findings indicate a clear answer to our research question posed earlier: yes, violent conflict affects self-employment. The nature of the effect of conflict on self-employment, however, varies across sectors (agriculture and services) and across measures of conflict (net displacement and attacks). Furthermore, this general finding has to be judged in the context of the available information on conflict and the case of Colombia. In the final paragraphs, we will comment on both issues in turn.

First, in terms of external validity, our analysis may be relevant for other middle-income, Latin American countries experiencing long-standing conflict. The external validity may apply especially to countries with similar labor market institutions. In essence, we find that the informal labor markets in Colombia respond quite flexibly to the burden of violence (in contrast to some Sub-Saharan African output markets in the face of conflict). This may be encouraging for similar countries experiencing similar conflicts. If conflicts are not too extreme, economic institutions can adjust and markets can provide some degree of resilience and protection for individuals. Having said that, it remains to be studied in more detail how policies can increase labor market resilience in the face of conflict.

Second, we employ specific measures of conflict. Yet, it is likely that different individuals and households with different socioeconomic and geographic endowments experience conflict differently (Verwimp, Justino, and Brück 2009; Brück et al. 2010). While conflict in Colombia has been a constant feature of the political life of the nation in aggregate, at the local and individual levels this may not be the case. Our results already indicate that findings differ by region, perhaps reflecting nuanced differences in types of conflict experienced. It remains to be explored to what extent different types of attacks (e.g., massive vs. targeted attacks, attacks by different groups and deaths vs. injuries) have different effects on outcomes of interest. Hence, future work may wish to account better for the effects of conflict experiences on self-employment at the micro level to build on our findings.

Footnotes

Acknowledgments

We thank Wim Naudé, Fabio Sánchez, Valentina Calderón, and two anonymous referees as well as participants at the UNU-WIDER Project Workshop on Entrepreneurship and Conflict (Ulster, March 2009), the 25th European Economic Association Meetings (Glasgow, August 2010), the Third Doctoral Research Seminar in Development Economics (Berlin, September 2010), at the 6th HiCN Workshop (Bogotá, December 2010), and at the Global Economic Costs of Conflict Workshop (Berlin, January 2011) for helpful comments. We are grateful to Ana Maria Ibáñez and Fabio Sánchez (Universidad de los Andes) for giving us access to important data sets. The usual disclaimers apply. Teodora Boneva provided exceptional research assistance at an early stage of the project.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: Bozzoli acknowledges financial support from MICROCON, a five-year research program funded by the European Commission.

Notes

References

Abadie

Gardeazabal

. 2003. “The Economic Costs of Conflict: A Case Study of the Basque Country.” American Economic Review 93:113–32.

Alesina

Perotti

. 1996. “Income Distribution, Political Instability, and Investment.” European Economic Review 40:1203–28.

Altonji

Card

. 1991. “The Effects of Immigration on the Labor Market Outcomes of Less-Skilled Natives.” In Immigration, Trade, and the Labor Market, edited by Abowd

John M.

Freeman

Richard B.

, 201–34. Chicago: University of Chicago Press.

Attanasio

Meghir

Vera-Hernandez

. 2004. “Baseline Report on the Evaluation of Familias en Acción.” IFS Reports, Institute for Fiscal Studies, London, UK.

Banerjee

Newman

. 1993. “Occupational Choice and the Process of Development.” Journal of Political Economy 101:274–98.

Bernal

Raquel

. 2009. “The Informal Labor Market in Colombia: Identification and Characterization.” Desarrollo y Sociedad 63:145–208.

Bernhardt

1994. “Comparative Advantage in Self-Employment and Paid Work.” Canadian Journal of Economics 27:273–89.

Borjas

George

. 1994. “The Economics of Immigration.” Journal of Economic Literature 32 (4): 1667–717.

Borjas

Bronars

Trejo

. 1992. “Assimilation and the Earnings of Young Internal Migrants.” The Review of Economics and Statistics 74:170–75.

10.

Borjas

Freeman

Katz

. 1996. “Searching for the Effect of Immigration on the Labor Market.” American Economic Review 86:246–51.

11.

Borjas

Grogger

Hanson

. 2010. “Immigration and the Economic Status of African-American Men.” Economica 77:255–82.

12.

Brück

Avdeenko

Justino

Verwimp

. 2010. “Identifying Conflict and Violence in Micro-Level Surveys.” Report for the German Federal Ministry for Economic Cooperation and Development, Bonn, Germany.

13.

Bundervoet

Verwimp

Akresh

. 2009. “Health and Civil War in Rural Burundi.” Journal of Human Resources 44:536–63.

14.

Calderón

Ibáñez

. 2009. “Labor Market Effects of Migration-Related Supply Shocks: Evidence from Internal Refugees in Colombia.” Documentos CEDE No. 2009-14, Universidad de los Andes, Bogotá, Colombia.

15.

Camacho

Rodriguez

. 2010. Firm Exit and Armed Conflict in Colombia. Working Papers WP2010-94, World Institute for Development Economic Research (UNU-WIDER, Helsinki, Finland).

16.

Card

David

. 1990. “The Impact of the Mariel Boatlift on the Miami Labor Market.” Industrial and Labor Relations Review 43 (2): 245–57.

17.

Chiswick

1978. “The Effect of Americanization on the Earnings of Foreign-born Men.” Journal of Political Economy 86:897–921.

18.

Cramer

Hartog

Jonker

Van Praag

. 2002. “Low Risk Aversion Encourages the Choice for Entrepreneurship: An Empirical Test of a Truism.” Journal of Economic Behavior & Organization 48:29–36.

19.

Deininger

2003. Causes and Consequences of Civil Strife—Micro-Level Evidence from Uganda. Policy Research Working Paper Series 3045, The World Bank, Washington, DC.

20.

Demirguc-Kunt

Klapper

Panos

. 2009. Entrepreneurship in Post-Conflict Transition: The Role of Informality and Access to Finance. Policy Research Working Paper Series 4935, The World Bank, Washington, DC.

21.

Destré

Henrard

. 2004. “The Determinants of Occupational Choice in Colombia: An Empirical Analysis.” Cahiers de la Maison des Sciences Economiques bla04065a. Paris: Université Panthéon-Sorbonne.

22.

Eren

Sula

. 2009. The Effect of Ability on Young Men’s Self-Employment Decision: Evidence from the NELS. Working Papers 0913, University of Nevada, Department of Economics, Las Vegas.

23.

Evans

Jovanovic

. 1989. “An Estimated Model of Entrepreneurial Choice under Liquidity Constraints.” Journal of Political Economy 97:808–27.

24.

Fairlie

2002. “Drug Dealing and Legitimate Self-Employment.” Journal of Labor Economics 20:538–67.

25.

Gaviria

2002. “Assessing the Effects of Corruption and Crime on Firm Performance: Evidence from Latin America.” Emerging Markets Review 3:245–68.

26.

González

Villarreal

. 2006. More Pushed than Pulled: Self-Employment in Rural Mexico Ten Years after NAFTA. Working Papers 20063, Escuela de Graduados en Administración Pública y Políticas Públicas, Monterrey, Mexico.

27.

Grun

2008. Household Investment under Violence—The Colombian Case. Policy Research Working Paper Series 4713, The World Bank, Washington, DC.

28.

Haile

2008. Determinants of Self-employment in Urban Ethiopia: Panel Data Based Evidence. PSI Discussion Paper 1, Policy Studies Institute, London.

29.

Hamilton

2000. “Does Entrepreneurship Pay? An Empirical Analysis of the Returns to Self-Employment.” Journal of Political Economy 108:604–31.

30.

Hundley

2000. “Male/Female Earnings Differences in Self-Employment: The Effects of Marriage, Children, and the Household Division of Labor.” Industrial and Labor Relations Review 54:95–114.

31.

Ibáñez

Vélez

. 2008. “Civil Conflict and Forced Migration: The Micro Determinants and Welfare Losses of Displacement in Colombia.” World Development 36:659–76.

32.

IDMC (Internal Displacement Monitoring Centre). 2009. Colombia: New Displacement Continues, Response Still Ineffective. A Profile of the Internal Displacement Situation. Colombia: Internal Displacement Monitoring Centre.

33.

Jacobs

2007. “An Occupational Choice Model for Developing Countries.” Proceedings of the German Development Economics Conference, Göttingen 2007 15, Verein für Socialpolitik, Research Committee Development Economics.

34.

Johansson

2000. “Self-Employment and Liquidity Constraints: Evidence from Finland.” Scandinavian Journal of Economics 102:123–34.

35.

Justino

Verwimp

. 2006. Poverty Dynamics, Violent Conflict and Convergence in Rwanda. HiCN Working Papers 16, Households in Conflict Network, Sussex, UK.

36.

sKan

Tsai

. 2006. “Entrepreneurship and Risk Aversion.” Small Business Economics 26:465–74.

37.

Kondylis

2010. “Conflict Displacement and Labor Market Outcomes in Post-War Bosnia and Herzegovina.” Journal of Development Economics 93:235–48.

38.

LaLonde

Topel

. 1991. “Labor Market Adjustments to Increased Immigration.” In Immigration, Trade, and the Labor Market, edited by Abowd

John M.

Freeman

Richard B.

, 167–200. Chicago: University of Chicago Press.

39.

Leibovich

Nigrinis

Ramos

. 2006. “Caracterización Del Mercado Laboral Rural En Colombia.” Borradores de Economia 408. Banco de la República de Colombia.

40.

Martínez

1998. “Colombia: La Medición de los Ingresos del Empleo Asalariado en las Encuestas de Hogares.” Taller 2 MECOVI. CEPAL, Santiago de Chile.

41.

Mesnard

2009. Migration, Violence and Welfare Programmes in Rural Colombia. IFS Working Paper W09/19, Institute for Fiscal Studies, London.

42.

Mesnard

Ravallion

. 2001. Is Inequality Bad for Business: A Non-Linear Microeconomic Model of Wealth Effects on Self-Employment. Policy Research Working Paper Series 2527, The World Bank, Washington, DC.

43.

Parker

2004. The Economics of Self-Employment and Entrepreneurship. Cambridge, MA: Cambridge University Press.

44.

Perfetti

2009. “Crisis y Pobreza en América Latina: El Caso de Colombia.” Manuscript, Rimisp, Santiago de Chile.

45.

Perry

Maloney

Arias

Fajnzylber

Mason

Saavedra-Chanduvi

. 2007. Informality: Exit and Exclusion. Washington, DC: The World Bank.

46.

Pshisva

Suarez

. 2010. “Capital Crimes: Kidnappings and Corporate Investment in Colombia.” In The Economics of Crime: Lessons for and from Latin America, edited by Di Tella

Edwards

Schargrodsky

, 63–97. Chicago: University of Chicago Press.

47.

Restrepo

Spagat

Vargas

. 2003. The Dynamics of the Colombian Civil Conflict: A New Data Set. CEPR Discussion Papers 4108. London: Centre for Economic Policy Research

48.

Rodríguez

Sánchez

. 2009. “Armed Conflict Exposure, Human Capital Investments and Child Labor: Evidence from Colombia.” Documentos CEDE 005400, Universidad de los Andes, Bogotá, Colombia.

49.

Shemyakina

2006. The Effect of Armed Conflict on Accumulation of Schooling: Results from Tajikistan. HiCN Working Papers 12, Households in Conflict Network. Sussex, UK.

50.

Stock

Yogo

. 2005. “Testing for Weak Instruments in Linear IV Regression.” In Identification and Inference for Econometric Models: Essays in Honor of Thomas J. Rothenberg, edited by Stock

J. H.

Andrews

D. W. K.

, 80–108. Cambridge University Press. http://ebooks.cambridge.org/ebook.jsf?bid=CBO9780511614491.

51.

Taylor

1996. “Earnings, Independence or Unemployment: Why Become Self-Employed?” Oxford Bulletin of Economics and Statistics 58:253–66.

52.

Verwimp

Justino

Brück

. 2009. “The Analysis of Conflict: A Micro-Level Perspective.” Journal of Peace Research 46:307–14.

53.

Van der Sluis

van Praag

Vijverberg

. 2005. “Entrepreneurship Selection and Performance: A Meta-Analysis of the Impact of Education in Developing Economies.” World Bank Economic Review 19:225–61.

Self-employment and Conflict in Colombia

Abstract

Keywords

Related Literature

A Theoretical Model

The Data

Descriptive Statistics: Displacement and Labor Market Transitions 7

Estimation Strategy

Conclusion

Footnotes

Acknowledgments

Declaration of Conflicting Interests

Funding

Notes

References

Descriptive Statistics: Displacement and Labor Market Transitions⁷