Скачать 43.94 Kb.
Контент2. Primary sources and data discussion
3. Methods of human capital evaluation in their application to the FSU
4. The spread of human capital in the FSU
Human capital in the former Soviet Union area:
sources and methods for the time-series construction*
Dmitry Didenko1, Péter Földvári2, and Bas van Leeuwen3
Human capital theory was put forward within neoclassical mainstream [Mincer, 1958; Schultz, 1961; Becker, 1962]. Their seminal works considered human capital as an individual stock of knowledge and skills, which tended to result in higher income levels of more educated individuals. In the 1980s and 1990s, with New Growth Theories [e.g. Lucas, 1988; Romer, 1990; Mankiw, Romer and Weil, 1992], the theoretical insight became widely accepted that human capital is the main source of long-run economic development. However, for the former socialist countries very little information on this factor is available.
In this paper we therefore develop a new and consistent dataset on human capital and related measures for the USSR and the Newly Independent States (NIS) after its dissolution. The dataset is a work in progress and will be located on the web-site of the Centre for Global Economic History at Utrecht University (http://www.cgeh.nl/human-capital-hub).
We construct the data series of various human capital indicators (both in natural- and monetary units), basically on an annual basis stretching back in most cases to 1920s. We add population (which is a crucial variable in many human capital estimates) in age-cohort breakdown, as well as comparable macroeconomic indicators like GDP, fixed capital stock, size of the general government expenditures, and the total wage bill, as well as price indices. Our focus is in constructing a dataset as clear, transparent and consistent as possible.
2.1 General description of the sources
Our starting point in constructing the dataset is the official statistics, available datasets and the research literature based on them.
As it was pointed out in [Davis and Wheatcroft, 1994] as well as in other literature starting at least from [Gershenkron, 1947], the Soviet official series contain the information that wasn’t intentionally falsified as the government statistical office preferred either to not to publish the unpleasant data or to adjust the methodology to let the resulting figures look better.
Among the major official series we rely primarily on the yearbook ‘The national economy of the USSR’ compiled by the Soviet Statistical Office (of various establishments) and on its topical volumes like ‘Labour’, ‘Construction of culture’, ‘Culture, education and science’, and ‘Females and children’. Likewise we used the general national budget execution reports of the Ministry of Finance and its specialised volumes on educational-, cultural services-, and research expenditures. For population data we depended on the results of 9 population censuses in the former Soviet Union area (hereinafter referred to as the FSU) and 2 in post-Soviet Russia. Additionally we employed some sources published by the Russian Empire, the CIS, and the Russian Federation Statistical Offices.
As for secondary sources, we use the data from [Maddison, 2010; National Research University Higher School of Economics, various years; UIS UNESCO, 2011; World Bank, 2011].
The research literature used includes, but is not limited to: [Becker, 1969; Bergson, 1961; Gregory, 1982; De Witt, 1961; Easterly and Fischer, 2001; Harrison, 1998; Johnson, 1950; Khanin, 1991; Markevich and Harrison, 2011; Mironov, 1985, 1994, 1996, 2003; Moorsteen and Powell, 1966; Noah, 1966; Plotnikov, 1954; Ponomarenko, 2002; Steinberg, 1990; Subbotina, 1965].
2.2. Population, its literacy and numeracy
We have assured that the data from [HSE IDEM, 2011] comply with those from the published census volumes except just minor cases. The discrepancies within the data for 1897, 1926, 1937 (most of all) and 1939 censuses are not considered as significant.
In all the FSU censuses literacy was defined as an ability to read at least one language therefore writing skills were not taken into account at all. In our opinion, conventional measurement based on direct questions left much room for reading proficiency criteria also to be eased, especially since literacy was a politically sensitive topic.
Innumeracy (age heaping) is measured as the excess of people reporting their ages ending on multiples of -5 and -0 (i.e. 25, 30, 35 etc). This measure is then converted into the ABCC index, proposed by [A’Hearn et al., 2009], which captures the percentage of persons correctly reporting their ages. Availability of the census data on 1-year age cohorts for male and female population at age 23-62 allows calculating their levels of numeracy, which is probably less upward-biased than literacy.
2.3. Educational attainment and enrolment
Educational attainment and enrolment from the sources was expressed for 6 ISCED levels for males, females and the total population separately.
In most cases we assign to each education level those durations of education that were normatively prescribed as of the census date. We assign the average value of the closest completed levels to duration of an in-between incomplete one.4 We assign the average duration of our detailed categories to a census-based broad education-level group.5
The major problem in operating with the Soviet-era enrolment series is their lack of full coverage of various types of educational establishments. We use both series on total enrolment and available incomplete series on education levels to estimate the complete ones, predominantly for the inter-war period (1920/21-1940/41 school years).
We use the resulting enrolment data to estimate educational attainment in years between censuses by following [Barro and Lee, 2001] perpetual inventory method. However, contrary to Barro and Lee, we took the average of forward- and backward-flow estimates as proposed by [Földvári and Van Leeuwen, 2009]. Applying the methodology as suggested by [Krueger and Lindahl, 2001], we get an error to signal variance ratio of 56% in our series versus 270% in [Barro and Lee, 2010]. This suggests that our series seem to outperform theirs by a considerable margin.
2.4. Financial data on human capital expenditures
To estimate expenditures for education proper we often use broader category ‘enlightenment’, which in the Soviet official financial reporting also included cultural services and, in certain periods, research.
The educational financial data are much better represented for the entire USSR than for its republics. Therefore we use the former to estimate the latter when it is necessary. Another our approach is estimating the share of a republic in total expenditures and then calculating absolute numbers. Logarithmic transformation is sometimes used for periods of high inflation (end 1920s-1930s, 1990s).
For the Soviet era we allocate the USSR central government budget between the republics. The size of consolidated budget of a particular republic is chosen as a single criterion to define its weight among the others in expenditures of the USSR central government.
We make allowance for the border changes in 1929 when Tajikistan split off from Uzbekistan and in 1936 when Kazakhstan and Kyrgyzstan split off from Russia.
2.5. Book production
Book production is often thought to be indicative of the accumulation of existing knowledge, e.g. in [Eisenstein, 1979]. Number of copies may be considered as a proxy for human capital quantity while number of titles for its quality. They may be a relatively reliable predictor of human capital before the ICT revolution (until 1990s in the FSU).
2.6. Labour market (employment and wages)
In order to valuate human capital, i.e. to determine how much a certain amount of schooling is worth, one needs information on the labour market.
Except for the period of mass compulsory labour during (and some time before, and after) WWII, a typical Soviet worker had substantial freedom of choice as to what education to obtain and what occupation to choose. That applied less to wage setting though. In our opinion, in the centrally-planned Soviet economy wage proportions were defined and set by the government. However, they were set to address shortage or abundance of particular skills and therefore affect their supply and demand primarily.
We used gross wages6 for blue- and white-collar workers on the observation that they were representative for wage development in general. The Soviet ruling elite considered industrial sector as the key one in the national economy. That is why the relationship between the wages of blue- and white-collar industrial workers may be considered as the core of the overall income distribution and, since, as a reliable proxy for human capital private returns.
The major weak point of the average wage official data is that they are upward-biased (especially from 1930s to 1960s) because of statisticians’ preference of industrial sector in terms of employment, while agricultural wages were significantly lower. Scarce official data appear from 1940 on employment and wages in collective agricultural enterprises for the USSR so that direct calculation of unbiased average wage becomes possible but only for selected years. For the FSU republics except Russia we have unbiased average wage data only from mid-1980s. To address this problem we used a retropolation correcting for the change in urban/rural population ratio.
2.7. National accounts (GDP, fixed capital) and price indices
Epistemologically, the Soviet official Net Material Product (NMP) concept was based on belief that no new value added may be created outside sectors of material production. Therefore, the official national income figures omitted most of services until mid-1980s. For that reason we take the GDP (GNP) estimations from the literature but also use the series of both NMP and GDP in current prices for their cross-check. We additionally verify estimations of the USSR GNP by the monetary indicators, like size of the general government expenditures and total wage bill, expressed in current prices. We have chosen to splice those series that had generally the same concepts and close values in time points to be linked together.
Our gross fixed capital stock estimation (in current prices) is based on gross fixed capital to GNP (at factor cost) ratio derived from [Easterly and Fischer, 2001] assuming this relationship, regardless of its monetary expression, is correct for any particular year. Like annual gross fixed investment, the accumulated stock values did not include those in livestock, inventories but did include those in residential housing, capital repairs in construction and installation services.
Our preferable inflation indicator is^ r as the most comprehensive price index that covers an entire economy. However we use consumer price indices for its construction and cross-check.
A problem in using the Soviet official statistics that estimated ‘real’ growth rates with earlier years as base (1913 or 1926/27) was identified in [Gerschenkron, 1947]. After deriving the price indices with different year base (1928, 1937, 1950, 1958, 1964, 1973, 1982) we address the so called ‘Gerschenkron effect’ by making our synthetic deflator where weights (e.g. of 1928 and 1937) are to change when they approach or diverge from the respective year base.
We test our Chained Deflator Index (CDI) for 1928-1955 by its application to the 1928 average wage, as of the Soviet official sources, following with comparison between actual and theoretical values. We find the resulting difference not substantial. To additionally check the adequacy of our CDI we construct a ratio of average wage to GDP per capita index with 1928 as our benchmark. Its dynamics generally fits the trends reported in the literature.
We compare the resulting real GDP growth rates with those derived from [Maddison, 2010] for the USSR (1928-1990). The discrepancy might arise from different deflator base. Maddison could deflate the GDP with retail price indices that evidently exceeded the entire GDP deflators about twice in 1930-1940s. Maddison could also ignore the data that demonstrated deflation in 1950-1955 and overall price stability in 1956-1958, which was assumed by us.
Such natural indicators like literacy, numeracy, book production numbers and volumes, and average years of education are surely not human capital proper but rather its proxies. However, in our case they may well be used to verify the monetary indicators or to go back in time where input monetary data are too scarce.
In calculating the cost-based measure, we follow [Judson, 2002], updated by [Van Leeuwen and Földvári, 2008], to value its stock per worker at replacement cost:
where ht denotes the average human capital stock per worker in year t, St is the average years of formal education in year t, djt is the public expenditure on education per level j in year t, ajt denotes the share of the labour force (population at the age of 15+ in the FSU case) in year t with a certain level of education.
This method does not include foregone wages and non-government spending on education. However, it is based on the key component of schooling costs, which normally defines their dynamics.
To employ income-based approach we follow the method proposed by [Van Leeuwen and Földvári, 2011]. Human capital is then treated in parallel with investments: the price of an asset, like a bond or a stock, will tend to be the present value of all expected future flows of income from it. Since, the present value of the future expected labour income of a worker, assuming continuous time and his/her retirement age at 65, can be expressed as:
where is per worker stock of human capital in monetary units, is actual average wage, is the average age in the population, g is constant rate of expected real wage growth and q is the discount factor. We assume that g-q=0.02, as people expect their utility resulting from higher wages will increase with time, in line with [Dagum and Slottje, 2000] at micro-level.
This method does not eliminate one’s earnings that would have been earned without any schooling. Therefore its results are not affected by intra-country wage differentials. However, if we assume that earnings of unschooled workers were equal across the FSU republics at any point in time, then their average wage differences would indicate the rewards for education, both private and social.
We identify the long-term trends of human capital spread in the FSU using various indicators.
Such proxies as literacy and numeracy display logistic pattern: after acceleration of growth it slows down as their values approach 100% (Figure 1).
However, if almost everyone can read and count, still people may acquire more schooling and benefit from it. As shown in Figure 2, the other series keep on moving about in the same direction.
Clearly, the income-based measure is strongly influenced by abrupt real wage dynamics. Notably, the former fluctuations tend to move reversely with those of white/blue-collar wage differential in industry (Figure 3). This highlights the pattern where positive social returns to education are gained in much at the expense of private ones.
In its turn, the cost-based measure is determined by education spending. Its extended relative indicator (public and private direct expenditures to GDP) correlated with real GDP p.c. level positively until end 1940s, while this pattern reversed afterwards (Figure 4). That could help to explain high economic growth rates in 1930-1960s, their slowdown and collapse thereafter. In 2000s positive relationship has restored, at least in Russia.
Gender gaps in ISCED 1-5 enrolment were steadily bridged during inter-war period with parity by 1940 in the USSR on average. However, male predominance persisted in the highest postgraduate level. Central Asian republics achieved gender parity in ISCED 1-3 enrolment only in 1960s, Turkmenistan didn’t ever report it as regards any higher level of education and some of these republics always had male predominance in tertiary enrolment.
Despite growth in number of book copies, the decline in their titles in 1960-1980s (from 335.2 to 292.2 per million inhabitants) is notable as indirect evidence of deterioration in the USSR human capital quality. However, the same indicator not merely recovered in Russia, but is at historical high at present (902.0 in 2009). This suggests that diversity of knowledge flows, even leaving electronic media aside, may have gotten a boost under open market system.
Human capital indicators kept growing in all of the FSU republics but unequally, so the republics could change their positions (Maps 1-3). E.g., Russia appeared to be the loser in all 3 rankings while Central Asian republics advanced in average years of education. However, inequality remained about equal in terms of income-based valuation, suggesting there was no catch up (Table 1). This is similar as was noticed by [Van Leeuwen, Van Leeuwen-Li and Földvári, 2011] in China, where low-income provinces failed to converge with rich ones.
Table 1: Human capital inequality in the FSU (Gini index)
Map 1: Income-based HC per worker (borders of 1989)
Map 2: Cost-based HC per worker (borders of 1989)
Map 3: Average years of education (borders of 1989)
The development of human capital in the FSU has been quite remarkable in an international perspective as well, especially comparing it with China (Figure 5).
Possibly, the faster spread of education in the USSR in 1920-1970s is one of the reasons it outperformed China in GDP p.c. growth during that period. However, in both countries periods of fast growth of human and physical capital were also the periods with the highest negative TFP growth.
In this paper, we develop a new dataset on human capital and related indicators for the FSU area, most of them between ca. 1920 and 2000. This fills a gap in the literature since so far very few estimates of this vast area have been made available.
We find that the FSU increased its human capital fast in the most part of the twentieth century, in much similar way as China did. This also applies for the particular republics, with too limited integration among them. Also some of the indicators provide evidence on deterioration in human capital spending level and its quality during late Soviet era, as well as on a few promising signs of their recovery in 2000s.
We anticipate that these results will stimulate further research into the role of human capital in the FSU/NIS development in global comparative context.
* The findings, interpretations, and conclusions are the authors’ own views and should not be attributed to the institutions of their affiliation. The authors acknowledge the financial support from the Netherlands Organisation for Scientific Research (NWO) under the CLIO-INFRA project.
1 Senior Analyst, State corporation ‘Bank for Development and Foreign Economic Affairs (Vnesheconombank)’ (Russia), firstname.lastname@example.org.
2 Associate Professor, University of Debrecen (Hungary), email@example.com.
3 Postdoc Researcher, Utrecht University (the Netherlands), firstname.lastname@example.org.
4 E.g., 8 years of ‘complete lower secondary’ and 10 of ‘complete upper secondary’ result in 9 for ‘incomplete upper secondary’.
5 E.g., 8 years of ‘complete lower secondary’ and 9 of ‘incomplete upper secondary’ result in 8.5 for ‘people with lower secondary education’.
6 Some data on blue- and white-collar workers were omitted in the sources. We estimate them based on total employment and average wage in the state-owned sector. In some cases (mainly for 1920s) we use time-series retropolation.