I analyze long-run trajectories of city growth using Abadie & Imbens (2002) bias-corrected matching estimator of average treatment effects for all three groups of factors: location relative to WWII front, industry evacuation, and Gulag. The choice of matching technique versus regression analysis allows me to remain agnostic about the functional form of the relationship between control variables and city growth. In principle, this relationship can be very unorthodox. The most obvious example of this is that the relationship between geographical variables and city growth does not have to be linear, or even have a simple functional form. Without a theory to guide the choice of a functional form, and with a large sample, matching estimators are a logical choice.
For each treated city, one or several matches from the control group are found. Matching cities must be as similar as possible to the treated city, with the similarity defined over the set of chosen characteristics. Then, the differences between treatment group and matching control group are analyzed in order to determine the effect of treatment.
To properly assess the effects of war, evacuation, or Gulag, we must compare cities of
similar administrative status and geographical location. As panel estimations show, the cities affected by WWII were different from the control group, even before the war. Therefore, I must also control for prior city growth and size. In all of the regressions, I exactly match cities on oblast center status. That is, I compare oblast centers with other oblast centers, and ordinary cities with other ordinary cities. Oblast centers may be more attractive than the average city because of their administrative functions.12
Cities also are matched on latitude, longitude, initial population level, and the rate of growth in the preceding time period. The algorithm looks for the closest matches in this four-dimensional space of matching characteristics (standardized by the sample variance), where the metric is given by the Euclidian distance (for details, see Abadie, Drukker, Herr
& Imbens (2001)).
Finally, for each of the treatments, the cities are matched exactly on other treatments.
For example, the growth of cities on both sides of WWII front lines is compared for those with the same evacuation status (whether enterprises were evacuated from the city, or to the city) and controlling for having a Gulag camp nearby. In the same way, cities that received evacuated plants are compared to cities that also were unaffected by WWII fighting, and to cities with the same Gulag status (presence or absence of a camp in the vicinity).
Table 7 presents the results for the WWII treatments. The effect of occupation is negative in 1939-1959 (as expected), but by 1970 the recovery is complete. There is no evidence that being occupied has any effect on city growth 25 years after the end of the war. This is in line with the results by Davis & Weinstein (2002), who found a similarly complete recovery from wartime destruction for Japanese cities.
12Another potentially important factor for the city growth are restrictions on residential mobility in the USSR. I borrow the data on residential restrictions (total and expansion restrictions) from Gang & Stuart (1999). Total restrictions were meant to prohibit all in-migration except for the cases of family reunion.
Expansion restrictions set targets for new labor from outside of a city that can be attracted by resident enterprises, and supposedly presented a weaker barrier for city growth. I tried matching cities on mobility restrictions, but the quality of matching is poor (not too many cities in the sample were restricted, it is difficult to find a match in the geographical vicinity). Yet the results with this control and without it were practically the same, which is an indirect confirmation of their result: administrative restrictions did not have a significant impact on city growth. I do not report these estimates in the paper.
Dependent variable isLnP opt−LnP opt−1
(1) (2) (3)
Treatment variable 30 km to 200 km to
Occupied front front # of obs
Time period
population 1939, population 1939, population 1939, growth 1926-1939, growth 1926-1939, growth 1926-1939, exact matching on oblast center oblast center oblast center
status, status, status,
evacuation (to, from), evacuation (to, from), evacuation(to, from), Gulag in 50 km. Gulag in 50 km. Gulag in 50 km.
Number of matches - 3, estimators are bias-adjusted for non-exact matching. Standard errors are heteroskedasticity- robust, ** denotes significance at 95% level, * - at 90% level.
Table 7: WWII and city growth, matching estimations.
As for evacuation status (Table 8), positive effects are observed for cities that received evacuation, but the effect is not statistically significant beyond 1970. Interestingly, the coef-ficient for treatment effect does not diminish over time, but its variance grows significantly.
Perhaps to fully explore the heterogeneity in the impact of evacuation, the researchers would need to collect more detailed archival data on the number of evacuees, size, or specialization of the enterprises evacuated.
Restricting matching destination-cities to the cities in the same macro-region does not change the results (column (2)). Cities, where industrial establishments did not return after the war, grew slower than the average until 1959, but the effect is short-lived and disappears in 1970 and later (column (5)).
As in the panel estimations, the positive and long-lasting effects for cities that sent plants into evacuation (column (3)) are due to the same statistical artifact of Soviet regional prior-ities and selection. When the matching algorithm places too much weight on geographical proximity, I end up pairing industrial cities in central Russia, where enterprises were evac-uated from, with cities in the same region where there was no important industry. When I do not match by longitude, this effect disappears (column (4)).
In contrast, the presence of a Gulag camp has a long-lasting and positive impact on city growth (table 9, column (1)). The difference between treatment and control group does not diminish, it actually continues to grow, even up to the present time! City size indices for all three treatments are plotted on Figure 11.13
In columns (2)-(5), of Table 9, I report the estimates of the treatment effects for dif-ferent types of Gulag camps. Camps that specialized in agriculture/forestry (most of these were logging operations, where prisoners worked) or construction had relatively weaker effect on cities. In “agriculture/forestry” camps, prisoners were used mainly to extract valuable resource (timber), not to create infrastructure for future development. Among the
“con-13Using “Gulag in 20 km” indicator produces even stronger results. As a robustness check, for the war and evacuation treatments I also did matching on the prior growth and size of city in 1926 (to have the same matching set as for Gulag treatment), the results are essentially the same. I do not report these results in the paper, but they are available upon request.
Dependent variable is LnP opt−LnP opt−1
(1) (2) (3) (4) (5)
Treatment variable enterprises enterprises enterprises enterprises enterprises
evacuated evacuated evacuated evacuated did not # of obs
to to from from return
Time period
1939-1959 0.109** 0.109** 0.378** -0.063* -0.106** 624
(0.036) (0.0367) (0.037) (0.038) (0.047)
1939-1970 0.120* 0.120* 0.262** -0.005 -0.051 625
(0.062) (0.062) (0.041) (0.043) (0.055)
1939-1979 0.114 0.113 0.188** 0.004 -0.056 625
(0.073) (0.073) (0.044) (0.046) (0.060)
1939-1989 0.122 0.120 0.246** -0.013 -0.041 629
(0.082) (0.082) (0.049) (0.050) (0.073)
1939-2002 0.117 0.117 0.173** -0.025 -0.041 629
(0.084) (0.084) (0.051) (0.053) (0.72)
1939-2010 0.101 0.099 0.137** 0.021 -0.038 627
(0.085) (0.085) (0.052) (0.101) (0.068)
% of obs treated 25 25 16 16 8
% of exact matches 99 62 63 63 63
Matching variables latitude, latitude, latitude, latitude, latitude, longitude, longitude, longitude,
population population population population population
1939, 1939, 1939, 1939, 1939,
growth growth growth growth growth
1926-1939, 1926-1939, 1926-1939, 1926-1939, 1926-1939, exact matching on oblast oblast oblast oblast oblast
center center center center center
status, status, status, status, status, war front war front war front war front war front in 200 km, in 200 km, in 200 km, in 200 km, in 200 km,
Gulag in Gulag in Gulag in Gulag in Gulag in
50 km. 50 km, 50 km. 50 km, 50 km,
Urals, longitude. longitude.
Siberia.
Number of matches - 3, estimators are bias-adjusted for non-exact matching. Standard errors are heteroskedasticity- robust, ** denotes significance at 95% level, * - at 90% level.
Table 8: Wartime enterprise evacuation and city growth, matching estimations.
Dependent variable is LnP opt−LnP opt−1
Gulag camp in 50 km
(1) (2) (3) (4) (5)
Treatment variable resource agriculture and
all camps extraction industry forestry construction # of obs Time period
1926-1939 0.104** 0.126* 0.223** 0.134** 0.097** 459
(0.046) (0.076) (0.061) (0.057) (0.048)
1939-1959 0.076** 0.218** 0.161** 0.060 0.090** 458
(0.031) (0.053) (0.033) (0.048) (0.031)
1926-1959 0.177** 0.323** 0.381** 0.190** 0.185** 458
(0.061) (0.094) (0.074) (0.085) (0.062)
1926-1970 0.176** 0.279** 0.387** 0.162* 0.200** 458
(0.073) (0.085) (0.088) (0.085) (0.073)
1926-1979 0.198** 0.321** 0.416** 0.195** 0.224** 458
(0.082) (0.093) (0.099) (0.089) (0.084)
1926-1989 0.222** 0.381** 0.422** 0.213** 0.216** 459
(0.088) (0.089) (0.104) (0.092) (0.095)
1926-2002 0.217** 0.355* 0.412** 0.214** 0.215** 459
(0.091) (0.096) (0.109) (0.094) (0.095)
1926-2010 0.226** 0.398** 0.455** 0.273** 0.231** 458
(0.093) (0.095) (0.111) (0.099) (0.098)
% of obs treated 46 18 24 13 34
% of exact matches 97 91 95 93 97
Matching variables latitude, latitude, latitude, latitude, latitude, longitude longitude longitude longitude longitude population population population population population
1926, 1926, 1926, 1926, 1926,
growth growth growth growth growth
1897-1926, 1897-1926, 1897-1926, 1897-1926, 1897-1926,
exact matching on oblast oblast oblast oblast oblast
center center center center center
status, status, status, status, status,
evacuation evacuation evacuation evacuation evacuation (to, from), (to, from), (to, from), (to, from), (to, from), war front war front war front war front war front in 30 km in 30 km in 30 km in 30 km in 30 km
Number of matches - 3, estimators are bias-adjusted for non-exact matching. Standard errors are heteroskedasticity- robust, ** denotes significance at 95% level, * - at 90% level.
Table 9: Gulag and city growth, matching estimations.
struction” camps were those created for the infamous infrastructural projects of the 1930s:
Northern Railroad, White-Sea-Baltic canal. Some of these projects proved a failure and were abandoned.
On the other hand, camps that specialized in industrial production (either primary in-dustries or other manufacturing) were creating this coveted “eastern industrial base” of the Soviet Union. Their impact on city size is twice as strong as that of an average camp.
Figure 11: City size index, as implied by matching estimations.
Table 10 presents estimated treatment effects for the different types of construction in Gulag. Consistent with Table 9, construction of industrial objects and housing leads to stronger long-run population increases. Construction of infrastructure shows weaker effects.
7 Conclusion
It is well understood that Gulag made possible many investment projects in the far corners of the USSR. I show that it also brought significant and long lasting changes to the spatial economy of the Soviet Union. Its impact worked on both the interregional and the intra-regional scale.
Dependent variable isLnP opt−LnP opt−1
Gulag camp in 50 km
(1) (2) (3) (4)
Treatment variable industrial industrial
construction construction housing infrastructure
(primary) (manufacturing) construction construction # of obs Time period
1926-1939 0.206** 0.097* 0.160** 0.105** 459
(0.062) (0.056) (0.067) (0.049)
1939-1959 0.248** 0.141** 0.146** 0.082** 458
(0.065) (0.036) (0.041) (0.031)
1926-1959 0.460** 0.236** 0.302** 0.184** 458
(0.093) (0.072) (0.086) (0.063)
1926-1970 0.399** 0.227* 0.290** 0.212** 458
(0.094) (0.079) (0.094) (0.076)
1926-1979 0.412** 0.255** 0.318** 0.243** 458
(0.109) (0.087) (0.102) (0.087)
1926-1989 0.393** 0.274** 0.333** 0.232** 459
(0.108) (0.092) (0.106) (0.095)
1926-2002 0.389** 0.282** 0.336** 0.226** 459
(0.115) (0.096) (0.109) (0.097)
1926-2010 0.443** 0.337** 0.392** 0.241*** 458
(0.120) (0.100) (0.117) (0.101)
% of obs treated 11 17 21 31
% of exact matches 90 95 94 97
Matching variables latitude, latitude, latitude, latitude, longitude longitude longitude longitude population population population population
1926, 1926, 1926, 1926,
growth growth growth growth
1897-1926, 1897-1926, 1897-1926, 1897-1926, exact matching on oblast center oblast center oblast center oblast center
status, status, status, status,
evacuation evacuation evacuation evacuation (to, from), (to, from), (to, from), (to, from), war front war front war front war front
in 30 km in 30 km in 30 km in 30 km
Number of matches - 3, estimators are bias-adjusted for non-exact matching. Standard errors are heteroskedasticity- robust, ** denotes significance at 95% level, * - at 90% level.
Table 10: Construction by Gulag prisoners and city growth, matching estimations.
To what extent was Gulag responsible for the reallocation of productive resources toward the remote regions of the USSR? Millions of people went through the Gulag system, but compared to the Soviet population, the size of Gulag labor force does not not seem eco-nomically significant. Even at its maximum, the able-bodied population of Gulag did not exceed 2% of the Soviet labor force (Khlevnyuk (2004)). A vast majority of Gulag camps were located close to population centers, where the size of the local labor force significantly exceeded the size of the camp population. But the presence of a camp is a good indicator of a local deficit of labor. Gulag was not the only tool of Soviet regional industrial policy, but its presence is a good signal that the location was chosen by the Soviet authorities for investment projects. What we observe in the data is probably not the effect of Gulag per se, but rather the combined effect of Soviet location policy for which Gulag is a good proxy.
Gulag camps were heterogeneous. Some of the camp locations were oriented exclusively toward resource extraction, were not planned as permanent settlements, and quickly withered after Stalin’s death. We know examples of abandoned camps in the middle of empty frozen landscapes. But the camps that were located close to the existing population centers were used to build basic industrial and public infrastructure, and to supply labor for industrial facilities, a part of long-term regional planning strategies. Such locations continue to attract population even after the Gulag system (and prison labor in general) has stopped functioning as a source of slave labor.
The effect of Gulag is much stronger than the estimated effects of WWII or reallocation of wartime industry. WWII is an example of exogenous impact, although Soviet authorities partially relocated productive resources away from the western border in preparation for it.
Evacuation was designed by the Soviet authorities, but it was done under the pressure of the Nazi invasion and thus should have served the purpose of maximizing Soviet industrial potential in wartime, i.e. in the short run. In contrast, Gulag was a part of long-run Soviet location policy, it was deliberately planned, and it served long-term goals. The changes to the Soviet spatial economy landscape in the 1930s to 1950s proxied by Gulag were perceived
as (and indeed were) permanent. My findings provide yet another illustration of “successful”
regional policy. It a strong suggestion that to be able to change economic geography in the long run, the impact of regional policy has to be as significant as Stalin’s industrialization of the eastern USSR.
The strongest long-term effect found here is for Gulag camps that specialized in industry, industrial construction, and construction of housing. What are the mechanisms behind this?
Do Stalin-era investments in capital and infrastructure still make the cities attractive, or are there other history-dependent factors? Is there a difference in local industrial structure between cities with Gulag camps and cities without them? What about sectoral diversity and specialization? Is there a difference in human capital? I leave these questions for further research.
References
Abadie, A., Drukker, D., Herr, J. L. & Imbens, G. W. (2001), ‘Implementing Matching Estimators for Average Treatment Effects in Stata’, The Stata Journal 1, 1–18.
Abadie, A. & Imbens, G. W. (2002), Simple and Bias-Corrected Matching Estimators for Average Treatment Effects, NBER Technical Working Papers 0283, National Bureau of Economic Research, Inc.
Andrienko, Y. & Guriev, S. M. (2004), ‘Determinants of Interregional Mobility in Russia’, Economics of Transition 12(1), 1–27.
Andronikov, N. G., ed. (1975), Istoria Vtoroi Mirovoi Voiny 1939-1945 (History of WWII 1939-1945), Vol. 4, Voenizdat, Moscow.
Applebaum, A. (2003),GULAG: A History, Doubleday.
Bosker, M., Brakman, S., Garretsen, H. & Schramm, M. (2007), ‘Looking for Multiple Equilibria When Geography Matters: German City Growth and the WWII Shock’, Journal of Urban Economics 61(1), 152–169.
Brakman, S., Garretsen, H. & Schramm, M. (2004), ‘The Strategic Bombing of German Cities during World War II and its Impact on City Growth’, Journal of Economic Geography 4(2), 201–218.
Davis, D. R. & Weinstein, D. E. (2002), ‘Bones, Bombs, and Break Points: The Geography of Economic Activity’, American Economic Review 92(5).
Davis, D. R. & Weinstein, D. E. (2008), ‘A Search for Multiple Equilibria in Urban Industrial Structure’, Journal of Regional Science48.
Dexter, K. & Rodionov, I. (2012), ‘The Factories, Research and Design Establishments of the Soviet Defence Industry: a Guide: Version 13’, University of Warwick, Department of Economics, http://warwick.ac.uk/vpk/.
Dienes, L. (1972), ‘Investment Priorities in Soviet Regions’, Annals of the Association of American Geographers 62(3), 437–454.
Dyker, D. A. (1983), The Process of Investment in the Soviet Union, Cambridge University Press, Cambridge, UK.
Gang, I. N. & Stuart, R. C. (1999), ‘Mobility Where Mobility is Illegal: Internal Migration and City Growth in the Soviet Union’,Journal of Population Economics12(1), 117–134.
Harris, C. D. (1970), Population of Cities of the Soviet Union, 1897, 1926, 1939, 1959 and 1967 : with tables, maps, and gazetteer, American Geographical Society.
Harrison, M. (1988), ‘Resource Mobilization for World War II: the U.S.A., U.K., U.S.S.R., and Germany, 1938-1945’, Economic History Review41, 171–192.
Harrison, M. (1990), ‘Stalinist Industrialisation and the Test of War’, History Workshop Journal (29), 65–84.
Hooson, D. J. (1968), The Growth of Cities in Pre-Soviet Russia, in R. P. Beckinsale &
J. M. Houston, eds, ‘Urbanization and its Problems: Essays in Honor of E. W. Gilbert’, Oxford: Basil Blackwell.
Huzinec, G. A. (1977), ‘Reexamination Of Soviet Industrial Location Theory’, The Profes-sional Geographer 29(3), 259–265.
Ivanova, G. M. (2006), Istoria Gulaga 1918-1958 / History of Gulag 1918-1958, Nauka, Moscow.
Iyer, S. D. (2003), ‘Increasing Unevenness in the Distribution of City Sizes in Post-Soviet Russia’, Eurasian Geography and Economics 44(5), 348–367.
Khlevnyuk, O. V., ed. (2004), Istoria Stalinskogo Gulaga / History of Stalin’s Gulag, Vol. 3 (Economika Gulaga / The Economy of Gulag), ROSSPEN.
Kozlov, M., ed. (1985), Velikaya Otechestvennaya Voina 1941-1945. Entsiklopediya. / The Great Patriotic War. An Encyclopedia., USSR Ministry of Defence, Institute of Military History and Sovetskaya Entsiklopediya, Moscow.
Krugman, P. (1991), ‘Increasing Returns and Economic Geography’, Journal of Political Economy 99.
Kumo, K. (2006), Interregional Population Migration in Russia: Using an Origin-to-Destination Matrix, Discussion Paper Series a483, Institute of Economic Research, Hitotsubashi University.
Lieberman, S. R. (1983), ‘The Evacuation of Industry in the Soviet Union during World War II’, Soviet Studies 35, 90–102.
Miguel, E. & Roland, G. (2011), ‘The Long-run Impact of Bombing Vietnam’, Journal of Development Economics 96(1), 1–15.
Redding, S. J. & Sturm, D. M. (2008), ‘The Costs of Remoteness: Evidence from German Division and Reunification’, American Economic Review 98(5), 1766–97.
Redding, S. J., Sturm, D. M. & Wolf, N. (2011), ‘History and Industry Location: Evidence from German Airports’, The Review of Economics and Statistics93(3), 814–831.
Rodgers, A. (1974), ‘The Locational Dynamics Of Soviet Industry’,Annals of the Association of American Geographers 64(2), 226–240.
Smirnov, M. (1998),Sistema Ispravitel’no-Trudovyh Lagerei v SSSR/The System of Correc-tive Labor Camps in the USSR, Zvenia, Moscow.