Constructing County-Level Data for Agricultural Inputs and Analyzing Agricultural Productivity, 1951-1980

Since the liberation from the Japanese occupation in 1945, South Korea has achieved substantial improvement in the nutritional status of the population, as indicated by the increase in adult heights. Recent studies suggest that increase in local food availability was an important contributing factor of the increased heights of the individuals born prior to 1960. Besides its significance as a long-term factor of improvement in nutritional status, measuring agricultural productivity and determining its major factors in the 1960s and 1970s is an important issue in its own right given the relative size of the Korean agricultural sector at the time. However, in-depth studies on agricultural productions in the past are restricted by the shortage of micro-level data covering the periods prior the 1980s.

In this study, I collected data sources (statistical yearbooks published by each province and county) and constructed databased containing variables regarding major inputs of agricultural productions in the 1960s and 1970s. I examined how major agricultural inputs (including land, labor, agricultural machines, and chemical fertilizers) changed over time and across provinces. By linking the data on inputs with the county-level agricultural production data, I also estimated agricultural production functions, focusing on the production of rice, the most important crop in Korean agriculture.

The present study is distinct from previous studies on Korean agricultural production in several respects. First, this research investigates agricultural production in Korean prior to 1980 based on county-level data, whereas most of previous studies that looked into the period are largely based on aggregate data of the country as a whole. Secondly, this study is the first to utilize the comprehensive county-level agricultural data on both outputs and inputs that are drawn from statistical yearbooks covering the two decades from 1960 to 1980. Finally, the present studies consider a wider range of agricultural inputs than those included in previous studies, including individual machinery and chemical fertilizer.

The area planted with all food crops and the size of rice-cultivating area increased and reached the peak in the mid 1965s. Afterwards, it declined over time. During the Korean War (1950 to 1953), the cultivated area temporarily diminished perhaps due to wartime destructions. The area of arable lands considerably differed by province. During the three decades under study, the province with the largest planted area was Gyeongbuk, followed by Jeonnam and Gyeongnam. By the 1970s, Jeonnam overtook Gyeongnam at the number one province in terms of the arable land area.

The farm population sharply fell from 1949 to 1951 as a consequence of wartime deaths. After the Korean War, the farm population gradually increased until 1967, and then declined over time thereafter. During the three decades under study, the top three provinces in terms of the size of farm population were Jeonnam, Gyeongbuk, and Gyeongnam. Even if the farm population is standardized according to age and gender compositions, these patterns of changes in labor input across times and provinces remain unchanged.

The number of major agricultural machines, such as power tillers, auto sprays, and tractors, increased sharply from the early 1970s. However, the trends should be cautiously interpreted because the relatively small number of machines in the early 1960s could result from the larger number of missing observations. Nevertheless, it seems evident that the availability of agricultural machines increased over time, although we cannot be sure how much under-reporting affects the real trend. If we compare years 1969 and 1980 when the number of counties with the number of machines reported remained unchanged, the number of power tillers increased more than 30 times. The increasing trend is similar to those of auto sprays and tractors. The patterns of changes in the use of agricultural machines substantially differed by region.

As in the case of agricultural machines, the use of chemical fertilizers dramatically increased from the early 1970s. Again, however, the trends should be taken cautiously because there were more counties in the 1960s where fertilizer consumption is unreported than in the 1970s. To address such potential problems, I also examined the yearly consumption divided by the number of counties (i.e., the average consumption per county). The results indicate that the rapid increasing trend largely captures the increase in the number of counties reporting fertilizer consumption. Furthermore, large fluctuations in each province’s fertilizer consumption are observed. These results suggest that samples with information on fertilizers should be selected so that variables for chemical fertilizers can be considered in the estimation of agricultural production functions.

Combining the county-level data on agricultural outputs and inputs, I estimated production functions of rice, the most important crop in Korean agriculture. The variable pertaining to land input is defined as the size the rice-cultivating area (measured in hectare) in each county in a given year. For labor input, I use the standardized population living in farm households cultivating rice. Since variables pertaining to capital inputs are not universally reported in provincial or county Annual Statistics, there is a tradeoff between considering more variables on inputs and additional loss of observations. I attempt to circumvent this problem in the following two ways. Firstly, I estimate agricultural production functions excluding the variable on capital inputs, and then extend the model by including additional capital inputs to examine the effects of the sample selections arising from missing observations of capital inputs. Secondly, I only focus only on major components of capital inputs to achieve a balance between omitted variables and missed observations. Finally, I included only the counties with information on a particular type of capital input (machine or fertilizer) to avoid bias arising from underreporting in early periods.

The results of regressions suggest that land and labor inputs have very strong positive relationship with the amount of rice production. In particular, the size of land input alone explains more than 95% of variations in rice productions across counties and years. If included separately, difference in labor input account for 83% of variations in rice outputs across counties and years. If the two inputs are included at the same time, the coefficient for land (0.99) is estimated much larger in magnitude than that for labor (0.05), confirming the huge importance of land in rice production in the 1960s and 1970s. If the year fixed is controlled, the coefficient for land diminishes by about 0.1 whereas the coefficient for labor increases by roughly the same magnitude. It is likely that year fixed effect captures the contributions of omitted factors that changed over time, including increased capital inputs and technical progress. The regression results imply that such omitted factors are positively related to land input, and negatively related to labor input. This is consistent with the fact that labor input decreased more rapidly than land input during the period under study.

I also conducted regressions in which a measure capital input (composite index of agricultural machines) is included. The coefficient for machine is positive and statistically significant, but the additional input explains only 3% of the variations in rice production across counties and years. If machine is additionally included, the coefficients for land and labor do not change much. Inclusion of year fixed effect reduces the coefficients for land and machine, whereas the contribution of labor becomes larger in magnitude. In particular, the coefficient for machine diminishes by more than two thirds. This indicates that the estimated contributions of agricultural machines largely capture the changes in capital input and output across times.

In sum, the results of regression analyses suggest that local rice production in Korea during the period from 1960 to 1979 was largely determined by land and labor inputs. Changes in these two factors explain more than 95% of variations in rice production across counties and years. It is difficult to estimate accurately the contributions of capital inputs to agricultural production because data are available only for selected capital inputs and for selected places and years. The results based on using three major agricultural machines of the time (power tillers, automatic sprays, and tractors) suggest that capital inputs also played significant roles in changing agricultural production, especially across times.

Given the currently available county-level data on agricultural inputs, it would be reasonable to use the number of major agricultural machines as an index of capital input in estimating agricultural production function. Land, labor, and agricultural machines explain over 98% of the variations in rice production across counties and years. Using the data and estimated regression coefficient for each input, it will be possible to estimate the agricultural total factor productivity as well as each factor productivity in each county and year. I remain it as future research agenda to investigate how natural, institutional and technological factors (such as natural disasters, local organizations, and new methods) affected these measures of local agricultural productivity.