Diffusion of (Non-)Discriminatory Culture: Evidence from Stalin s Ethnic Deportations

Transcription

1 : Evidence from Stalin s Ethnic Deportations Alain Blum (EHESS) Alexandra Jarotschkin (PSE-EHESS) Ekaterina Zhuravskaya (PSE and EHESS) December 2017

2 Intro Background Data Results Conclusions Motivation Deportations Question Literature Motivation When people get exposed to a different culture, do they reject it or adopt it? Do cultural traits diffuse or does exposure to a new culture increase people s identification with their own? An ideal experiment designed to answer this question would assign people of different cultural backgrounds randomly to the same locations The literature has studied random allocation of children to classes, refugees to social housing, students to dorms, and soldiers to regiments Such experiments imply a controlled environment, in which representatives of different cultures are forced to interact

3 Intro Background Data Results Conclusions Motivation Deportations Question Literature Motivation Yet, in a real-life setting, where people choose freely whom to interact with, they may self-segregate and avoid representatives of a different culture There are many examples of spontaneously-created ghettos both in history and throughout the world along ethno-religious lines in 19th century European towns racial lines in contemporary American cities religious lines in contemporary European cities To study cultural diffusion, one needs to combine an experimental setting of cultural exposure with having no control over whom people interact with Stalin s ethnic deportations during WWII combine both of these features

4 Intro Background Data Results Conclusions Motivation Deportations Question Literature Stalin s ethnic deportations 2.1 million people were deported from the Western parts of the USSR to Siberia and to the Central Asian republics in With the sole reason of belonging to an ethnicity suspected of (potential or actual) collaboration with Germans against the Soviets Deportees were not allowed to leave the settlements (until Khrushchev s thaw or the fall of the USSR, depending on ethnicity) but they were not confined to camps, unlike Gulag s prisoners which left the possibility of interaction with local population

5 Intro Background Data Results Conclusions Motivation Deportations Question Literature Culture of deportees Culturally, deported ethnicities differed along many dimensions, e.g., in terms of religion: Protestants: Germans, Estonians, Latvians, and Finns Muslims: Chechens, Ingush, Karachays, Balkars, Kurds, Crimean tatars, Turk-meshketians Catholic christians: Poles, Lithuanians Orthodox christians: Moldovans, Greeks Buddhists: Kalmyks, Koreans In particular, gender norms sharply differed between Muslim deportees from North Caucasus and Protestant deportees from the North-West Chechens and Volga Germans were the largest groups of Muslims and Protestants, respectively We focus on the effect of deportations on this cultural trait

6 Intro Background Data Results Conclusions Motivation Deportations Question Literature Research question What are the gender norms among the native local population today in Siberia and Central Asia depending on whether they live in a locality that was exposed to Muslim or Protestant ethnic deportations in the past The vast majority of the descendants of the deportees left when they were allowed to And we look at the native local population Preview: we find strong evidence of cultural diffusion

7 Intro Background Data Results Conclusions Motivation Deportations Question Literature Related literature Social identity and cultural diffusion literature focus on the effect of co-existence of ethnic and racial groups In non-experimental setting (Tajfel and Turner, 1979; Grosfeld et al., 2013; Sakalli, 2015) In (quasi-)experimental setting (Algan, Hemet and Laitin 2016 on social housing; Chetty Hendren 2015, Rao 2013 on classroom composition; Burns et al on dorms assignments; Angrist, 1998; Vanden Eynde, 2011 on military service) Determinants of gender roles (Fernandez and Fogli 2009; Fernandez, Fogli and Olivetti 2004) Effects of ethnic deportations on distrust in central authority (Levkin 2015, using crude region-level data)

8 Intro Background Data Results Conclusions Timing Size Destination Gender attitudes Timeline of ethnic deportations Deportations: : deportations from annexed territories, Poland, Baltic republics, and Romania : Preventive deportations of Soviet Germans, Finns, and Greeks : Retributive deportations of the ethnic groups of the Northern Caucasus and Crimea Deportees were allowed to return to their homelands in two waves: 1991: Volga Germans, Crimean Tatars, Turk-Meskhetians 1956: All the rest

9 Intro Background Data Results Conclusions Timing Size Destination Gender attitudes Deportations by religion and the destination s Soviet republic (in 000s of people) Soviet republic of destination All Russia Kazakhstan Uzbekistan Kyrgyzstan Tajikistan Protestants 1, Muslims Catholics Greek Orthodox Eastern Orthodox Total 2,

10 Intro Background Data Results Conclusions Timing Size Destination Gender attitudes Deportation locations Soviet republic of destination All Russia Kazakhstan Uzbekistan Kyrgyzstan Tajikistan Districts (rayons) with deportations

11 Intro Background Data Results Conclusions Timing Size Destination Gender attitudes Central Asia and Eastern Siberia: size and religious composition of deportation settlements

12 Intro Background Data Results Conclusions Timing Size Destination Gender attitudes Check on data: NKVD archives vs Soviet census by region (oblast) in USSR Nb of Protestants as of 1970 Soviet Census ( ln ) Nb of Protestants, 1970 (ln) Fitted values Nb of Deported Protestants ( ln ) p value: Nb of Sunni Muslims as of 1970 Soviet Census ( ln ) Nb of Sunni Muslims, 1970 (ln) Fitted values Nb of Deported Sunni Muslims ( ln ) p value: By 1970 Chechens were pardoned and left but Germans were at the deportation locations

13 Intro Background Data Results Conclusions Timing Size Destination Gender attitudes Gender role norms of Muslims and Protestants There are no systematic data on gender norms at the time of deportations Evidence of polygyny with men having up to five wives in Chechen-Ingush ASSR 1963 (e.g., Ro i 2000, p. 539) Yet, there are some pieces of evidence from before and after Literacy gap between men and women was smaller for Protestants (and Germans) than for Muslims (and North Caucasians) in 1897: Rural literacy rates, 1897 Census Protestants Muslims Diff-in-diff: Germans North-Caucasians Diff-in-diff: Women Men Diff: Men-Women Urban literacy rates, 1897 Census Protestants Muslims Diff-in-diff: Germans North-Caucasians Diff-in-diff: Women Men Diff: Men-Women

14 Intro Background Data Results Conclusions Timing Size Destination Gender attitudes Gender role norms of Muslims and Protestants Contemporary survey data (Life in Transition survey, 2016): Protestants Muslims Significance of difference Agree that a woman should do most of the household chores 26% 67% *** Agree that it is better for everyone if the man earns the money 51% 73% *** Agree that men make better political leaders 41% 67% *** Membership in a women group 4% 4% Tertiary education, among women 14% 10% *** Tried to start a business, among women 11% 8% *** Obs. 2,053 12,567 A lot of anecdotal evidence that religious and political leadership in today s Chechen republic is not opposed to pre-arranged marriages of teenage girls with middle-age men, polygyny, and female genital mutilation

15 Intro Background Data Results Conclusions Timing Size Destination Gender attitudes The largest deported Protestant group Volga Germans Catherine II called on Europeans to immigrate to Russia in the late 18th century promising religious freedom, exemption from military service, and thirty years without having to pay taxes Volga German colonists were refugees from the war-ravaged German states The bulk of those Germans came from largely protestant regions of Hesse and Palatinate Settlers were allowed to keep their language, traditions, and religion

16 Intro Background Data Results Conclusions Timing Size Destination Gender attitudes Gender equality was the official policy of USSR Female genital mutilation, polygamy, child marriage were forbidden in the Soviet Union Equality for men and women was proclaimed as part of ideology, including in the sphere of education Liquidation of illiteracy campaigns in the 1920s and 30s targeted equally men and women All people from 8 to 50 years old were required to become literate in their native language Boys and girls had the same schooling obligations It is hard to say whether native population of Central Asian republics was less backward than Chechens at the time of deportations However, there is evidence that the deported North-Caucasian groups resisted the Soviet ideology, including regarding gender equality, arguably more than Central Asian population (although there was resistance there too, i.e., Basmachi movement)

17 Intro Background Data Results Conclusions Sources Econometric Strategy Variation Balance Data sources 1 Ethnic deportations Alain Blum: data on exact destinations and the numbers of deportees for each ethnicity in each deported settlement NKVD Kommendatura census of all deportation locations made in 1951, Russian National Archives GARF 19,839 entries, 18 ethnic groups (Koreans are missing) non-ethnic deportees: Kulaks, bandits, anti-soviet elements we matched the destination locations of these entries with 1126 districts (municipalities in the USSR) 2 Contemporary outcomes Life in Transition survey data, PSUs in 5 countries Out of which 232 PSUs had a deportation in 30km vicinity 3 Historical and geographical controls Gulag locations Old and new capitals, railroads, ruggedness, water ways, etc.

18 Intro Background Data Results Conclusions Sources Econometric Strategy Variation Balance Summary statistics Sample: All Conditional deportation vicinity Mean SD Min Max Mean SD Min Max Nb of Protestant deportees Nb of Muslim deportees Share of Protestant deportees Share of Muslim deportees Size of deportations Capital (old or new) Urban Distance to capital (km) Distance to railroad (km) Distance to gulag (km) Distance to water (km) Ruggedness Observations

19 Intro Background Data Results Conclusions Sources Econometric Strategy Variation Balance Econometric Strategy We explore gender norms of local population in PSUs which were exposed to deportations of Protestant vs. Muslim groups Controlling for subnational region (oblast) fixed effects and a variety of historical and geographical characteristics Identification assumption: conditional on covariates and the presence of deportation in the vicinity, the identity of the deportees (e.g., their religion) was random Balancing tests (presented below) suggest that this is the case We restrict the sample to the title ethnicity of each country (i.e., Kazakhs in Kazakhstan; Russians in Russia) to make sure that descendants of deportees are not in the sample

20 Intro Background Data Results Conclusions Sources Econometric Strategy Variation Balance Econometric Strategy Y il = β 0 + β 1 log(p rotestants l ) + β 2 log(muslims l )+ + ω j log(others j l ) + γ X l + δ C i + µ rl + ɛ il Y il = α 0 + α 1 log(p rotestantshare l ) + ω j log(others j l )+ +γ X l + δ C i + µ rl + ɛ il i - respondents, i - localities (PSUs) log(p rotestantshare) l - The share of Protestant deportees historically present within a 30km vicinity of the locality l log(p rotestants) l, log(muslims) l, log(others j l ) - logs of the number of Protestant, Muslim and other deportees historically present within a 30km vicinity X and C - vectors of PSU-specific controls: past/present capital city, current urban/rural status, distances to railroads, nearest capital city, Gulag camp, and individual-level controls: education, age, income, gender, and religious group dummies µ rl - subnational region fixed effect Cluster by PSU

21 Intro Background Data Results Conclusions Sources Econometric Strategy Variation Balance The biggest variation among LiTs PSU exposure to ethnic deportations in Kazakhstan and Kyrgyzstan 56 PSUs in Kazakhstan and 62 PSUs in Kyrgyzstan had a deportation settlement with Muslim or Protestant groups Variation in Kazakhstan Variation in Kyrgyzstan percent percent Protestant share Muslim share Protestant share Muslim share PSUs that had a Muslim and/ or Protestant deportation in Kazakhstan PSUs that had a Muslim and/ or Protestant deportation in Kyrgyzstan

22 Intro Background Data Results Conclusions Sources Econometric Strategy Variation Balance Less variation, but still some, in Uzbekistan, and Tajikistan 59 PSUs in Uzbekistan and 31 PSUs in Tajikistan had a deportation settlement with Muslim or Protestant groups Variation in Uzbekistan Variation in Tajikistan percent percent Protestant share Muslim share Protestant share Muslim share PSUs that had a Muslim and/ or Protestant deportation in Uzbekistan PSUs that had a Muslim and/ or Protestant deportation in Tajikistan

23 Intro Background Data Results Conclusions Sources Econometric Strategy Variation Balance... and Russia 24 PSUs in Russia had a deportation settlement with Muslim or Protestant groups Variation in Russia percent Protestant share PSUs that had a Muslim and/ or Protestant deportation in Russia Muslim share

24 Intro Background Data Results Conclusions Sources Econometric Strategy Variation Balance Balance Deportation locations were not random, yet, the religious composition seems to be random Dep.Var.: Capital Dist. to railroad Ruggedness Dist. to water Dist. to capital Dist. to Gulag old or new (ln) (ln) (ln) (ln) (ln) Panel A: Sample of all PSUs Protestant deportees (ln) * (0.009) (0.048) (0.017) (0.043) (0.030) (0.030) Muslim deportees (ln) 0.019** * * ** * (0.009) (0.049) (0.018) (0.044) (0.030) (0.031) Observations Region Fixed Effects Yes Yes Yes Yes Yes Yes Panel B: Sample of PSUs with deportation in vicinity, no size control Share of protestant deportees ** (0.138) (0.598) (0.200) (0.554) (0.409) (0.390) Total deportation size NO NO NO NO NO NO Observations Panel C: Sample of PSUs with deportation in vicinity, with control for deportation size Share of protestant deportees (0.140) (0.593) (0.202) (0.562) (0.408) (0.382) Total deportation size Yes Yes Yes Yes Yes Yes Observations Region Fixed Effects Yes Yes Yes Yes Yes Yes

25 Intro Background Data Results Conclusions Attitudes Behavioral outcomes Time-variant effect A woman should do most of the household chores even if the husband is unemployed Do you agree? Protestant deportees (ln) *** ** (0.007) (0.009) Muslim deportees (ln) (0.007) (0.006) p-value: Protestants = Muslims 0.012** 0.015** Agree or strongly agree that: A woman should always do most of the household chores Share of Protestant deportees *** *** (0.054) (0.097) Observations 2,679 2,005 1,640 1,232 R-squared Other Deportations Yes Yes Yes Yes Demographic Controls Yes Yes Yes Yes Region FE Yes Yes Yes Yes Sample - Deportations in Vicinity All All Conditional Conditional Sample - Gender Female Male Female Male Mean of dependent var s.d. of dependent var

26 Intro Background Data Results Conclusions Attitudes Behavioral outcomes Time-variant effect It is better for everyone involved if the man earns the money Do you agree? Protestant deportees (ln) * ** (0.007) (0.008) Muslim deportees (ln) (0.006) (0.004) p-value: Protestants = Muslims 0.036** 0.037** Agree or strongly agree that: It is better if the man earns the money Share of Protestant deportees * (0.075) (0.090) Observations 2,656 1,996 1,633 1,230 R-squared Other Deportations Yes Yes Yes Yes Demographic Controls Yes Yes Yes Yes Region FE Yes Yes Yes Yes Sample - Deportations in Vicinity All All Conditional Conditional Sample - Gender Female Male Female Male Mean of dependent var s.d. of dependent var

27 Intro Background Data Results Conclusions Attitudes Behavioral outcomes Time-variant effect Men make better political leaders than women do Do you agree? Protestant Deportees (ln) *** * (0.007) (0.007) Muslim Deportees (ln) 0.012** (0.005) (0.005) p-value: Protestants = Muslims 0.000*** Agree or strongly agree that: Men make better political leaders than women do Protestant Deportees (Share) * ** (0.077) (0.075) Observations 2,635 1,979 1,618 1,211 R-squared Other Deportations Yes Yes Yes Yes Demographic Controls Yes Yes Yes Yes Region FE Yes Yes Yes Yes Sample - Deportations in Vicinity All All Conditional Conditional Sample - Gender Female Male Female Male Mean of dependent var s.d. of dependent var

28 Intro Background Data Results Conclusions Attitudes Behavioral outcomes Time-variant effect First principal component of gender attitudes from the survey all factor loadings are positive Protestant deportees (ln) *** *** (0.005) (0.007) Muslim deportees (ln) 0.011*** (0.004) (0.004) p-value: Protestants = Muslims 0.001*** 0.018** 1st principal component Gender attitudes Share of Protestant deportees ** *** (0.052) (0.063) Observations 2,572 1,925 1,596 1,187 R-squared Other Deportations Yes Yes Yes Yes Demographic Controls Yes Yes Yes Yes Region FE Yes Yes Yes Yes Sample - Deportations in Vicinity All All Conditional Conditional Sample - Gender Female Male Female Male Mean of dependent var s.d. of dependent var

29 Intro Background Data Results Conclusions Attitudes Behavioral outcomes Time-variant effect Are you a member of a women s group? Significant results for men Protestant deportees (ln) ** (0.003) (0.003) Muslim deportees (ln) (0.003) (0.003) p-value: Protestants = Muslims ** Are you a member of a women s group Share of Protestant deportees * (0.068) (0.063) Observations 2,732 2,048 1,666 1,251 R-squared Other Deportations Yes Yes Yes Yes Demographic Controls Yes Yes Yes Yes Region FE Yes Yes Yes Yes Sample - Deportations in Vicinity All All Conditional Conditional Sample - Gender Female Male Female Male Mean of dependent var s.d. of dependent var

30 Intro Background Data Results Conclusions Attitudes Behavioral outcomes Time-variant effect Did you try to start a business? Significant results for women, men only sample is placebo Protestant deportees (ln) 0.007** (0.003) (0.006) Muslim deportees (ln) *** (0.003) (0.006) p-value: Protestants = Muslims 0.003*** Did you try to start a business? Share of Protestant deportees 0.099** (0.045) (0.087) Observations 2,732 2,048 1,666 1,251 R-squared Other Deportations Yes Yes Yes Yes Demographic Controls Yes Yes Yes Yes Region FE Yes Yes Yes Yes Sample - Deportations in Vicinity All All Conditional Conditional Sample - Gender Female Male Female Male Mean of dependent var s.d. of dependent var

31 Intro Background Data Results Conclusions Attitudes Behavioral outcomes Time-variant effect Womens education and the educational gap parents and respondents generations Sample: individual individual individual individual PSU PSU Dep. Var.: Mother s tertiary Diff: mother-father Diff: female-male educ post-mandatory educ respondents tertiary educ Protestant deportees (ln) 0.007* (0.004) (0.003) (0.008) Muslim deportees (ln) * (0.004) (0.003) (0.008) p-value: Protestants = Muslims 0.034** Share of Protestant deportees * (0.042) (0.034) (0.102) Observations 3,975 2,383 3,898 2, R-squared Other Deportations Yes Yes Yes Yes Yes Yes Demographic Controls Yes Yes Yes Yes Yes Yes Region FE Yes Yes Yes Yes Yes Yes Sample - Deportations in Vicinity All Conditional All Conditional All Conditional Mean of dependent var s.d. of dependent var

32 Intro Background Data Results Conclusions Attitudes Behavioral outcomes Time-variant effect Coeff. on log(muslim deportees) X Cohort Did your mother obtain tertiary education? Quasi-Panel by birth cohorts of respondent: number of Protestant and Muslim deportees Mother s education is the only variable for which we observe pre-treatment differences We can use birth cohorts of respondents as time dimension in a quasi-panel estimation The power is not large, but the direction of differences as predicted Mother obtained tertiary education point estimates 95% confidence intervals Mother obtained tertiary education point estimates 95% confidence intervals.3.3 Coeff. on log(protestant deportees) X Cohort Pre Pre Birth cohort of respondent Birth cohort of respondent

33 Intro Background Data Results Conclusions Attitudes Behavioral outcomes Time-variant effect Did your mother obtain tertiary education? Quasi-Panel by birth cohorts of respondent: the share of Protestant deportees Mother obtained tertiary education Coeff. on protestant share X cohort Pre 1951 point estimates 95% confidence intervals Birth cohort of respondent

34 Intro Background Data Results Conclusions Attitudes Behavioral outcomes Time-variant effect Coeff. on log(muslim deportees) X Cohort The mother s education is not what is driving the results on attitudes of respondents All the results go through controlling for parents education, also, the attitudes effects are there irrespective of the cohort Attitude toward women, 1st principal component point estimates 95% confidence intervals Attitude toward women, 1st principal component point estimates 95% confidence intervals Coeff. on log(protestant deportees) X Cohort Pre Pre Birth cohort of respondents Birth cohort of respondents

35 Intro Background Data Results Conclusions Attitudes Behavioral outcomes Time-variant effect Permanent effect of the share of Protestant deportees on attitudes Gender attitudes, 1st principal component Coeff. on Protestant share X Cohort Pre 1951 point estimates 95% confidence intervals Birth cohort of respondent

36 Intro Background Data Results Conclusions Attitudes Behavioral outcomes Time-variant effect Larger deportees settlements matter more: Semi-parametric regressions 1st principal component Gender attitudes <1000 Protestant deportees * ** (0.046) (0.045) (0.057) (0.064) 1-10k Protestant deportees *** *** ** *** (0.056) (0.062) (0.065) (0.073) 10-20k Protestant deportees *** *** (0.082) (0.125) (0.097) (0.147) 20-30k Protestant deportees * ** * ** (0.060) (0.066) (0.073) (0.073) <1000 Muslim deportees ** (0.049) (0.047) (0.089) (0.110) 1-10k Muslim deportees 0.078* *** (0.047) (0.045) (0.097) (0.116) 10-20k Muslim deportees ** (0.064) (0.057) (0.119) (0.133) 20-30k Muslim deportees *** (0.083) (0.078) (0.126) (0.144) Observations 2,572 1,925 1,596 1,187 R-squared Other Deportations Yes Yes Yes Yes Demographic Controls Yes Yes Yes Yes Region FE Yes Yes Yes Yes Sample - Deportations in Vicinity All All Conditional Conditional Sample - Gender Female Male Female Male Mean of dependent var s.d. of dependent var

37 Intro Background Data Results Conclusions Attitudes Behavioral outcomes Time-variant effect Central Asia vs. Russia: the effects are there in both 1st principal component Gender attitudes Central Asia Protestant deportees (ln) *** ** (0.007) (0.008) Muslim deportees (ln) 0.012** (0.005) (0.004) p-value: Protestants = Muslims 0.002*** 0.055** Share of Protestant deportees ** (0.061) (0.073) Observations 1,972 1,574 1,367 1,076 R-squared Mean of dependent var Russia Protestant deportees (ln) * (0.007) (0.012) Muslim deportees (ln) 0.027*** (0.009) (0.020) p-value: Protestants = Muslims 0.001*** Share of Protestant deportees *** (0.085) (0.192) Observations R-squared Mean of dependent var p-value Protestant: Cent.Asia=Russia *** p-value Muslim: Cent.Asia=Russia Standard controls Yes Yes Yes Yes Sample - Deportations in Vicinity All All Conditional Conditional Sample - Gender Female Male Female Male

38 Intro Background Data Results Conclusions Attitudes Behavioral outcomes Time-variant effect Religiosity does not interact with treatment 1st principal component Gender attitudes Protestant deportees (ln) ** * (0.006) (0.008) Muslim deportees (ln) 0.012** (0.005) (0.004) Trust - religious institutions 0.073*** 0.057** (0.024) (0.024) (0.023) (0.025) Trust - religious institutions Protestant (ln) (0.006) (0.007) Trust - religious institutions Muslim (ln) (0.005) (0.005) Share of Protestant deportees ** (0.062) (0.090) Trust - religious institutions Protestant share (0.064) (0.082) Observations 1,758 1,373 1, R-squared Other Deportations Controls Yes Yes Yes Yes Demographic Controls Yes Yes Yes Yes Region FE Yes Yes Yes Yes Sample - Deportations in Vicinity All All Conditional Conditional Sample - Gender Female Male Female Male Mean of dependent var s.d. of dependent var

39 Intro Background Data Results Conclusions Conclusions We document the diffusion and persistence of a (non-)discriminatory culture We use Stalin s ethnic deportations as a historical experiment, in which the close co-existence of different cultural groups is imposed in a real-word setting without any constraints on the interaction between groups We find that both the norms of gender equality and gender discrimination can be adopted by a group if that group is exposed to another group with those norms Both the attitudes toward gender equality and the actual behavioural outcomes among the local population in historical locations of Stalin s deportations in Central Asia and Siberia depend on the gender norms of the deported group