Data
We use cross-sectional nationally representative data from the Life in Transition-III survey (LiTS-III),Footnote 3 collected by the European Bank of Reconstruction and Development and the World Bank in 2015/2016. The survey covered 29 post-socialist countries of CEE and Central Asia (including Mongolia), as well as Turkey, Greece, Cyprus, Germany, and Italy. Information on former Communist party membership was only collected in post-socialist states. The survey excluded Turkmenistan.
In each country, the LiTS-III conducted 1500 face-to-face interviews. Households were selected according to a two-stage clustered stratified sampling procedure. In the first stage, the frame of primary sampling units was established using information on local electoral territorial units. In the second stage, a random walk fieldwork procedure was used to select households within primary sampling units. Further information about the survey design and implementation are available in the LiTS Annex (EBRD 2016).
Variables
Outcome variable: entrepreneurial activity
A unique feature of the LiTS-III is that it contains detailed information on respondents’ past and present entrepreneurial activities. Our dependent variable (started business) is based on responses to the question “Have you ever tried to set up a business?,” with possible answers “Yes, I have set up my current business,” “Yes, I set up a business in the past but I am no longer involved in it or it is no longer operational,” “Yes. I tried to set up a business and did not succeed (in setting it up),” and “No.”
Main explanatory variable: former communist party membership
Our key independent variable captures connections with the former Communist party based on whether respondents themselves, their parents, or other family members were party members prior to 1989/1991.Footnote 4 First, the variable any personal or family link to the former Communist party takes value 1 when the respondent has a personal or family connection with the former Communist party and 0 otherwise. Next, we created three separate dichotomous variables measuring the following: (i) individuals who themselves were party members; (ii) the children of former party members; and (iii) the relatives (other than children) of former party members. These categories can overlap because the respondent can be a former party member themselves and at the same time have parents or relatives who were also party members. About 21% of respondents in our analysis sample report links to the former Communist party, ranging from 39% in Montenegro to 12% in Hungary.
Control variables
Our regressions include standard control variables used in the entrepreneurship literature (e.g., Aidis et al. 2008; Block et al. 2015; Demirgüç-Kunt et al. 2007; Djankov et al. 2005; Estrin et al. 2013a, b; Nikolova and Simroth 2015). Specifically, we add to our regressions three sets of control variables. First, the set of individual- and household-level controls consists of the respondent’s gender, age and its square, ethnic minority status, religious affiliation, retirement and disability status, respondent’s height, respondent’s education, a wealth index based on information about household assets, employment status, marital status, household size, number of children under 18, subjective health assessment, risk attitudes, current membership of any political party, parental education, and the number of books at home during the respondent’s childhood. Second, the set of geography-related controls consists of the urbanity status (capital, urban-not-capital, rural), latitude, longitude, and elevation of the respondent’s place of residence. Finally, to account for all possible country-level influences and capture within-country relationships between former Communist party membership and entrepreneurship, we include country-fixed effects.Footnote 5
To avoid bias from dropping observations with missing information, we create an additional category for missing information where the share of missing observations for a particular categorical variable is greater than 1%. The only continuous variable with a share of missing observations higher than 1% is that of respondents’ height (11% missing observations); here we create within-country height tertiles and treat the variable as categorical with missing observations being the fourth category. The missing category for these variables has no particular interpretation but serves to preserve the number of observations. The summary statistics on all variables used in the analysis is available in Table S1 in the Supplementary Information file.Footnote 6
Estimation strategy
We model the entrepreneurship outcome started business of each individual i living in country j as follows:
$$ {\mathrm{Started}\ \mathrm{business}}_{ij}={\upbeta}_0+{\upbeta}_1\ {\mathrm{Communist}\ \mathrm{party}}_{ij}+{\boldsymbol{X}}_{ij}^{\prime}\boldsymbol{\gamma} +{\varepsilon}_{ij} $$
(1)
where Communist party captures personal or familial ties to the former Communist party, X is a vector of control variables described above, and ε is the stochastic error term. Given the likely interdependence of respondent outcomes at the local level, we cluster the standard errors at the primary sampling unit (PSU) level. Given the categorical and unordered nature of the dependent variable, we rely on a multinomial logit estimator. The underlying assumption of the multinomial logit model is the independence of irrelevant alternatives (IIA), i.e., the assumption that the relative probability of choosing between two options is independent of additional alternatives in the choice set. We have tested for the IIA assumption and found that in our case it is not violated.Footnote 7
The parameter β1 captures the association between former Communist party membership and entrepreneurial activity rather than a causal effect. The Communist party variable is potentially endogenous, meaning that β1 may not reflect the true causal effect of party membership on entrepreneurship but rather self-selection into entrepreneurship across households. For example, individuals living in households with certain family environments, or observed or unobserved characteristics related to motivation or ability, may be more likely to both be or be linked to a former party member and start a business.Footnote 8
To mitigate endogeneity issues and identify causal effects, we employ a control function approach, a technique that is similar to the instrumental variable approach (Petrin and Train 2010; Rivers and Vuong 1988; Wooldridge 2015), and suitable for non-linear models, such as the multinomial logit that we use in our analysis.Footnote 9 Sometimes referred to as a two-stage residual inclusion (2SRI) (Terza et al. 2008), this approach necessitates one or more variables—instruments—that are highly correlated with the endogenous regressor (former Communist party membership) and affect the outcome (entrepreneurship) only through the endogenous regressor. In the first stage, the exogenous variation brought by the instruments also induces variation in the generalized residuals, which serve as control functions (Wooldridge 2015). Including the control functions in the second stage renders the endogenous independent variable (former Communist party membership) plausibly exogenous. The advantage of the control function approach in our case is that it also handles non-linear endogenous variables.
Following Terza et al. (2008), we estimate a first-stage auxiliary regression using a generalized linear model, with a binomial family and probit link, whereby the potentially endogenous regressor (i.e., former Communist party membership) is explained by the instruments and all the control variables. Next, we include the predicted first-stage residuals, along with the endogenous regressor, in the second stage estimation. The standard errors in the second stage and the reported marginal effects are based on bootstrapped clustered replications. The coefficient estimate on the endogenous regressor in the second stage represents the unbiased effect of the former Communist party on entrepreneurial activity, while that on the predicted residuals captures the endogeneity bias. We estimate the following system of equations:
$$ {1}^{\mathrm{st}}\ \mathrm{stage}:{\mathrm{Communist}\ \mathrm{party}}_{ij}={\gamma}_0+{\gamma}_1{\mathrm{Instruments}}_{ij}+{\boldsymbol{X}}_{ij}^{\prime}\boldsymbol{\pi} +{u}_{ij} $$
(2)
$$ {2}^{\mathrm{nd}}\ \mathrm{stage}:{\mathrm{Started}\ \mathrm{business}}_{ij}={\overset{\sim }{\beta}}_0+{\overset{\sim }{\beta}}_1{\mathrm{Communist}\ \mathrm{party}}_{ij}+{\upalpha u}_{ij}^{est}+{\boldsymbol{X}}_{\boldsymbol{ij}}^{\prime}\overset{\sim }{\boldsymbol{\gamma}}+{\zeta}_{ij} $$
(3)
where, for each individual i in country j, X denotes a vector of all control variables (including country-fixed effects), u is the error term of the first-stage regression, uest is the predicted residual from the first-stage equation, and ζ is the error term in the second-stage regression.
One advantage of the control function approach is that the inclusion of the residuals provides a Hausman test of the null hypothesis that the Communist party variable is exogenous (Bollen et al. 1995; Wooldridge 2015). Specifically, the statistical significance of the coefficient estimate on the predicted residuals, denoted by α in Eq. (3), indicates that the Communist party membership variable is endogenous, implying that the control function rather than simple multinomial regression results should be used for interpretation.
Following Ivlevs and Hinks (2018), we instrument personal and family links to the former Communist party with information about the involvement of respondents’ family members in the Second World War (WWII). Ivlevs and Hinks (2018) summarize evidence showing that across the former socialist world, WWII veterans (and in many cases civilians who were affected by war) were encouraged and given priority to join the Communist party and take leading positions in the government and various administrative bodies. Our expectation is that people who themselves, or whose parents and grandparents, fought in, or were otherwise affected by, WWII would be more likely to have either personal affiliation or family links to the former Communist party (instrument relevance).
The assumption about instrument exogeneity, that being affected by WWII (or being the descendant of such people) is uncorrelated with the error term, necessitates further discussion. While this assumption is fundamentally untestable, we provide historical arguments about its plausibility. It is unlikely that when WWII started, people joined the army because they thought it would make it easier to join the ruling party during or after the war which, in turn, would help them to establish businesses after a regime change. However, some people might have wished to go to war in anticipation of joining the party and benefiting from the advantages that party membership brings (for example, be in power/managerial positions and exploit advantageous networks). If the personal characteristics of such individuals are linked with traits that determine the likelihood and success of entrepreneurial activities (Caliendo et al. 2014), the instruments may not be exogenous. To ensure that these considerations pose no threat to our instrument’s validity, following Ivlevs and Hinks (2018), we concentrate on the CEE countries that did not have communist regimes before WWII. This allows us to rule out the possibility that people in these countries joined the war effort in order to become members of the Communist party during or after the war. There are 18 such countries in our sample—Poland, Czech Republic, Slovakia, Hungary, Romania, Bulgaria, Albania, the seven successor states of Yugoslavia (Bosnia and Herzegovina, Croatia, Former Yugoslav Republic of Macedonia, Kosovo, Montenegro, Slovenia, and Serbia), the Baltic States (Estonia, Latvia, and Lithuania), and Moldova.Footnote 10 Our analysis sample therefore only includes this set of countries.
We measure respondents’, their parents’, and grandparents’ involvement in WWII using information from two survey questions: (i) “Were you, your parents or any of your grandparents physically injured or were your parents or any of your grandparents killed during WWII?” and (ii) “Did you, your parents or any of your grandparents have to move as a result of WWII?,” with possible answers “Yes” and “No.” We construct two binary variables, killed/injured in WWII and displaced as a result of WWII,Footnote 11 and expect both to be positively correlated with personal or family links to the former Communist party. We note that apart from fighting in WWII, these variables would also capture broader WWII effects on civilians. However, in many cases—for example, relocation to a labor camp or participating in an underground resistance movement—civilians affected by WWII would also receive preferential treatment after the war.