Public Health and Bathing in Late Imperial Russia: A Statistical Approach

Abstract

Going to the Russian bathhouse (bania) helped people preserve their health, at least according to common idioms and many medical doctors at the end of the nineteenth century and into the twentieth century. Yet the thousands of public bathhouses across the empire were often damp, dank, and overcrowded places conducive to the spread of infectious diseases. Were Russian Imperial bathhouses healthy or dangerous? Did they prevent and cure ailments or did they foster the spread of sickness and in turn increase mortality and incidence of disease? Qualitative reports from the period are inconclusive: some cite the benefits of bathhouses, others cite their threat to public health, and many note both. In this paper we use data from reports produced in the 1890s and early 1900s to undertake a statistical analysis of the health effects of public bathhouses in late Imperial Russia. To our knowledge, this is the first attempt to apply modern statistical techniques to questions of Russian health and hygiene in this period. We find no evidence to support doctors' claims that banias were beneficial to public health and instead find some evidence that banias were associated with higher incidence of disease.

Footnotes

• 1

  I. I. Illiustrov, Sbornik rossiiskikh poslovits i pogovorok (Kiev, 1904).

• 2

  On physicians in the era see Nancy M. Frieden, Russian Physicians in an Era of Reform and Revolution, 1856–1905 (Princeton, 1981).

• 3

  Modern medical claims about the importance of the bania for health can be traced back to Anotonio Sanches, O parnykh rossiiskikh baniakh: Poeliku spospeshestvuiut oni ukrepleniiu, sokhraneniiu i vozstanovleniiu zdraviia (St. Petersburg, 1779); the theme was picked up by Russian researchers in the second half of the nineteenth century in part in response to Western doctors’ newfound appreciation of the health benefits of bathing. See, for instance, Sergei Gruzdev, Miniral'nyi obmen pri russkoi bane (St. Petersburg, 1890), A. Fadeev, Materialy k ucheniiu o russkoi ban: Vliianie bani na usvoenie i obmen seru, fosfora i khlora (St. Petersburg, 1890), and F. I. Vetoshikov, K voprosu o vliianii russkoi bani na usvoenie zhirov pishchi i zdorovykh liudei (St. Petersburg, 1894).

• 4

  See, for example, V. Znamesnkii, O russkikh baniakh v gigienicheskom otnoshenii (St. Petersburg, 1861); S. Iu. Fialkovskii, “Materialy k voprosu o vliianii bani na zdorovyi i bol'noi glas’ cheloveka,”Vrach, 1881, no. 9:137–43; V. V. Godlevskii, Materialy dlia ucheniia russkoi ban' (St. Petersburg, 1883); A. F. Mendes, Bani v obshchedostupnom izlozhenii (St. Petersburg, 1902); and Vestnik obshchestvennoi gigieny i prakticheskoi meditsini, 1915, vol. 51 pt. 1:713–14.

• 5

  Doctors tended to emphasize the importance of public bathhouses for positive health outcomes and dismiss the utility of private banias. Access to the private banias was limited, the resources required in time to prepare them and in fuel to heat them meant that they were not consistently used, and most of them were built without chimneys (the so called “black baths” because of the soot that built up on the walls as the room filled with smoke.) All of this meant that, even when private banias were used, doctors suspected that they posed significant health risks. See N. A. Maliavko-Vysotkaia, Pol'za i vred krest'ianskoi bani (Pskov, 1887) and V. Dolzhenkov, “K voprosu ob ustroistve sel'skikh obshchestvennykh ban',”Trudy VII S″ezda Zamskikh Vrachei i Predstavitelei Zemstve Kurskoi Gubernii (Kursk, 1900), 229–44.

• 6

  On the role of doctors in addressing medical problems amidst the social transformations of the late-nineteenth century in Russia, and in the context of changing theories of etiology, see Frieden, Russian Physicians; and Charlotte E. Henze, Disease, Health Care and Government in Late Imperial Russia (London, 2011).

• 7

  Jeff Wiltse, Contested Waters: A Social History of Swimming Pools in America (Chapel Hill, 2007); Lawrence Wright, Clean and Decent: The Fascinating History of the Bathroom and the Water Closet (New York, 1960).

• 8

  A. G. Cross, “The Russian Banya in the Descriptions of Foreign Travelers and in the Depictions of Foreign and Russian Artists,”Oxford Slavonic Papers, no. 24 (1991): 34–59.

• 9

  Alain Corbin, The Foul and the Fragrant: Odor and the French Social Imagination (Cambridge, MA, 1986); Georges Vigarello, Concepts of Cleanliness: Changing Attitudes in France since the Middle Ages (Cambridge, England, 1988); Virginia Smith, Clean: A History of Personal Hygiene and Purity (Oxford, 2007).

• 10

  I. V. Eremeev, Gorod S.-Peterburg s tochki zreniia meditsinkoi politsii (St. Petersburg, 1897); Mathhias Roth, The Russian Bath: Published with a View to Recommend its Introduction into England for Hygienic as well as Curative Purposes (London, 1852).

• 11

  I. I. Verevkin, “O russkikh baniakh,”Arkhiv sudebnoi meditsiny 4 (1865): 25–31.

• 12

  K. V. Markov, Voiskovyi bani i prachechnyi (Kiev, 1900).

• 13

  I. A. Karvasovskii, Kratkii ocherk’ istorii ban’ i znachenie ikh v gigienicheskom i terapevicheskom otnosheniiakh (Kiev, 1884). In writing this, Karvasovskii was quoting a colleague, with whom he agreed.

• 14

  Eremeev, Gorod S.-Peterburg, 262–64, 512–17; The prominence of venereal disease among banshchiki clearly had to do with the fact that they could be hired out for sex by clients. Pollock deals with sex in the bathhouses in a manuscript in progress. For an excellent analysis of the role of banshchiki in same-sex activities see Dan Healey, Homosexual Desire in Revolutionary Russia: The Regulation of Sexual and Gender Dissent (Chicago, 2001).

• 15

  Maliavko-Vysotskaia, Pol'za i vred krest'ianskoi bani. For a clear sense of the medical dangers of the bania see V. I. Chugin, “Zametka o russkikh baniakh v sanitar'nom otnoshenii,”Vrach, 1880, no. 35:577–78, and no. 36:585–87.

• 16

  Alexander Martin, “Sewage and the City: Filth, Smell, and Representations of Urban Life in Moscow, 1770–1870,”Russian Review 67 (April 2008): 243–74.

• 17

  See, for instance, Mary Douglas, Purity and Danger: An Analysis of the Concepts of Pollution and Taboo (London, 2002); Warwick Anderson, “Excremental Colonialism: Public Health and the Poetics of Pollution,”Critical Inquiry (Spring 1995): 640–69; Kathleen M. Brown, Foul Bodies: Cleanliness in Early America (New Haven, 2009); and Timothy Burke, Lifebuoy Men, Lux Women: Commodification, Consumption, and Cleanliness in Modern Zimbabwe (Durham, 1996). Indeed, foreigners’ accounts of the bania before the nineteenth century—and after the middle of the twentieth century—tend to use the bania as a way of marking the Russians as different and inferior.

• 18

  Articles and books on the bania from the period consistently praised the institution in principle even as they criticized the conditions in most baths and chastized the customers and bathhouse workers for being unhygienic in their behaviors in the bathhouse. Doctors generally concurred that the public banias were an improvement on the private baths, and their advocacy on behalf of bathing centered on the health benefits of public banias (see footnote 5).

• 19

  Of course, the richness of the qualitative data is essential for answering questions about people's assumptions about health and hygiene, the place of the bania in Russian culture and tradition, the appeal of the bania to various sectors of the population, and the bania as a site of social and sexual interaction. Pollock addresses these and other issues in the history of the bania he is currently writing.

• 20

  For the purposes of this study we defined “banias” as publicly accessible bathhouses. These existed in cities and towns across the empire and were usually run for profit after a license was obtained from the municipal authorities. We exclude private, household bathhouses, which did not appear in the statistics and which quantitative data suggest were rarely accessible to the urban population we are targeting.

• 21

  Sanitarnoe sostoianie gorodov Rossiiskoi imperii 1895 godu: Otchet’ Meditsinskago Departamenta Ministerstva Vnutrennykh del' (St. Petersburg, 1899). The bathhouse data were reported for towns and cities of the empire. The 1897 population variable we used to standardize the bania counts measured the urban population and excluded the surrounding rural areas. We followed the categorization of the population into urban and rural groups used within the 1897 Census. Suburban or semi-urban settlements such as slobodi and posadi were usually excluded. However, official publications of the Interior Ministry warn that in practice, it was often hard to determine exactly where the town boundary ended and the rural population of the district (uezd) began. Moreover, in some cases, settlements right beyond the town walls were counted as rural and thus the rural population of many districts outnumbered the town population.

• 22

  Although the bania data were initially reported and gathered at the town level, we aggregated them to the level of eighty-seven regions (see Part I of the Appendix). We do this because most of our data for health outcomes and other variables (see below) were only reported at the regional level. Therefore, we are explicitly exploiting inter-regional variation in bania density and health outcomes to test for the presence of a relationship between the two.

• 23

  We create city-level public bania counts from the Sanitarnoe sostoianie. Bania count is imputed at the city-level using multiple imputation (see Part III of the Appendix) and then aggregated to regional level. At the regional level, we standardize bania count using the 1897 urban population, sourced from the empire-wide 1897 Census, N. A. Troitskii, ed., Pervaia vseobshchaia perepis’ naseleniia Rossiiskoi Imperii 1897 goda (St. Petersburg, 1905).

• 24

  Mortality is averaged annual all-cause mortality from 1902 to 1910. Incidence of each of the diseases is averaged annual incidence from 1902 to 1910. In the otcheti, tif is segmented into sypnoi tif (epidemic typhus), briushnoi tif (typhoid fever), vozvratnyi tif (relapsing fever), and indeterminate tif. Modern medicine is more clear on the distinction between relapsing fever (vozvratnyi tif) and typhus (sypnoi tif)—which are transmitted through human lice and ticks—and typhoid fever (briushnoi tif)—which is spread by salmonella enterica in water contaminated by human waste. Because the symptoms for these diseases were often the same and because the data also included a large number of “undetermined” types of tif, we refer to the whole category as “tif” in the paper even as we take these distinctions into account in the results and conclusions. When running regressions, we transform all the outcome variables using the natural logarithm to better approximate a normal distribution. See Part II of the Appendix for detailed definitions of these variables and the sources we used to measure them. authorities. We excluded private, household bathhouses, which did not appear in the statistics and which qualitative data suggest were rarely accessible to the urban population we are targeting.

• 25

  See, for example, V. V. Godlevskii, Materialy dlia ucheniia russkoi ban (St. Petersburg, 1883) and V. Izopolskii, Pol'za i vred bani i ikh kratkii istoricheskii ocherk' (Kiev, 1885).

• 26

  See Part VII of the Appendix for a correlation table between the main eleven dependent variables. Note that we did not find multicollinearity to be a major issue in our regressions when we looked at the variance inflation factors (vif() command in R). The only variable with a VIF above a commonly cited benchmark of 5 was percent Jews, but it still fell within an acceptable range of 5–10. Moreover, excluding the percent Jews variable from the model as a robustness check yields the same substantive results. To confirm the robustness of our model, we also ran tests using data for other factors that would capture some of the same broader influences on health and bania density as those we already included. Specifically, we ran models where we inserted the number of schools per capita and the number of factory workers per 1000 people as proxies for modernization, but neither substantively changed the results. We also ran models using the mean age instead of the median age, percent Slavic language speakers as opposed to percent Russian speakers, and share of foreign workers out of the total work force as an alternative proxy for ethnicity. Again, the results remained robust to these changes.

• 27

  We control for climate using dummy variables for “supra-regions,” which are sixteen groupings of regions (excluding the Finnish guberniias) determined by the Imperial Russian Geographic Society. In statistical parlance, these are termed “fixed effects.” Specifically, these are binary variables that take a value of 1 when a region is in a given supra-region, and 0 otherwise. The use of fixed effects controls for climate under the assumption that climate is invariant across all the regions within the same supra-region. Moreover, fixed effects can capture other idiosyncrasies of supra-regional groups that are shared by all regions within that group. More technically, fixed effects are equivalent to subtracting the respective supra-regional means from all the dependent and independent variables at the supra-regional level (de-meaning). Therefore, in the regression specification with fixed effects, our explanatory variables are accounting for variation of the mortality rate around its supra-regional group averages.

• 28

  Numerous epidemiological and economic studies have found a link between economic development and mortality. For an example of a cross-national economic study of the relationship between income and mortality see David Cutler, Angus Deaton and Adriana Lleras-Muney, “The Determinants Of Mortality,” Journal of Economic Perspectives 20:3 (2006): 97–120. We believe that given the absence of GDP statistics in this era and other data limitations, tax revenue per capita serves as an effective proxy for economic development (we make the assumption that the tax rate and state taxation capacity is consistent across the Russian Empire). We used people per house as a proxy for housing density. At the time doctors assumed that the greater the number of people living in a household the more likely it was for disease to spread (and mortality to rise). Physicians per 100,000 urban residents serve as a proxy for the capacity of the public health system. The share of workers in agriculture allows us to capture the degree of industrialization of a region. This is significant given that previous research has shown that the degree of industrialization is an important predictor of mortality. See S. J. Pocock, et al., “The British regional heart study: geographic variations in cardiovascular mortality and the role of water quality,”British Medical Journal (1980), no. 270:1243–49.

• 29

  There was occasional missing data at the regional level, which we did not impute. For example, Samarkand was missing mortality data, forcing us to exclude it from regressions where mortality is a dependent variable.

• 30

  Sanitarnoe sostoianie reported only 5 banias in Iakutskaia oblast.

• 31

  The data gathered from Sanitarnoe sostoianie show that all four towns in the Ural'skaia guberniia reported that they had no banias. This result was not due to either missing data or our imputation procedure—furthermore, qualitative data suggest it was not all that rare for urban areas not to have a bania.

• 32

  These mortality and morbidity statistics represent the average number of cases per 100,000 urban residents per year from 1902 to 1910.

• 33

  The variability statistics about these diseases may suggest more about the quality of recording and reporting in different regions of the empire than they do about the actual regional variation in the incidence of the disease. But given that these are the available statistics, we use them with caution and awareness of the uncertainty of reporting accuracy.

• 34

  For more on the geographic distribution of diseases and epidemics in the empire see K. G. Vasil'ev, Istoriia epidemii v Rossii (Moscow, 1960).

• 35

  The curve for these figures is drawn for the Moskovsko-Promyshlennaia group, although the relationship in each case is identical for all groups. The difference between the sixteen supra-regional groups is that the y-intercept (the baseline mortality rate) is different. Graphically, one can represent this by shifting the relationship drawn for Moskovsko-Promyshlennaia group up or down along the y axis. We use the Moskovsko-Promyshlennaia group as our example because it is among the groups for which we have the most town-level observations.

• 36

  Henceforth, we adopt the complete specification of the model with fixed effects as our preferred specification (see Table 4B).

• 37

  See footnote 26 for details of additional variables used. Results not presented, but available from authors upon request.

• 38

  Chugin, “Zametka.”

• 39

  Lack of filtering was a consistent complaint in doctors’ reports and appears regularly in the descriptions of the empire's banias in the Sanitarnoe sostoianie.

• 40

  For an argument for how bania use—along with other measures—helped eradicate diseases in the USSR in the 1940s see Donald Filtzer, The Hazards of Urban Life in Late Stalinist Russia: Health, Hygiene, and Living Standards, 1943–1953 (Cambridge, England, 2010), esp. 127–62.

• 41

  Specifically, we aggregated to 87 regions from 715 cities.

• 42

  Multiple imputation is the preferred statistical techique for imputing missing data in much of the statistical literature (see, for instance, R. J. A. Little and D. B. Rubin, Statistical Analysis with Missing Data [New York, 1987]). The assumption we are required to make is that the missing data are missing completely at random (MAR) conditional on the covariates we use for imputation. Multiple imputation was carried out using Amelia II package in R. See Matthew Blackwell, James Honaker, and Gary King, “Multiple Overimputation: A Unified Approach to Measurement Error and Missing Data,” July 24, 2010, available at http://www.mattblackwell.org/files/papers/merror-methods.pdf (last accessed July 9, 2012).

• 43

  All of the variables except banias were transformed using the natural logarithm for imputation. The number of banias was transformed using the square root for imputation. Following additional research, we incorporated 95 percent confidence intervals on the number of banias as priors for the following cities: Odessa 20–30 banias, Kiev 10–18, Tashkent 1–2, Kishinev 2–7, Astrakhan 4–7, and Saratov 5–8.

• 44

  That is, following the rich literature in public health and epidemiology, we conceptualize death and disease as a Poisson process, where d_it is the number of people that die in the unit i (region in our case) during time period t. The mortality rate is calculated as m_it=d_it/p_it, where p_it is the total population in unit i at time t. Under the assumption that the expected number of deaths, d_it, is not close to zero, we can model m_it according to a log-normal distribution. This implies that the natural logarithm of the mortality rate, ln(m_it), is normally distributed. For additional details see Federico Girosi and Gary King, Demographic Forecasting (Princeton, 2008). We also transform some of the explanatory variables in order to reign in outliers and impose a more normal distribution on the data. Specifically, we transform bania density using the square root, which accommodates the fact that a region may have no banias. We transform doctor density, people per house, and revenue per capita using the natural logarithm.

• 45

  For details of the SUR model see Ferdinand Alimadhi, Ying Lu, and Elena Villalon, “sur: Seemingly Unrelated Regression” (2007), in Kosuke Imai, Gary King, and Olivia Lau, “Zelig: Everyone's Statistical Software,” available at http://gking.harvard.edu/zelig. We carried out all estimation of the SUR model using the Zelig package in R. For more details on Zelig see Kosuke Imai, Gary King, and Olivia Lau, “Toward A Common Framework for Statistical Analysis and Development,”Journal of Computational and Graphical Statistics 17 (December 2008): 892–913.

• 46

  Specifically, seemingly unrelated regression allows for correlation of error terms across equations for a given observation. The model can still be estimated using ordinary least squares when considering just a single outcome.

• 47

  Therefore, we assume the errors are not correlated between regions. Unfortunately, the lack of extensive data does not permit us to model cross-sectional dependence more explicitly.