![Sample our Education journals, sign in here to start your access, latest two full volumes FREE to you for 14 days]({"type":"image" , "src" : "/sda/5932/TOC-Subject-Sample-2021-Education.jpg"})
ABSTRACT
Purpose
This paper is concerned with the impact of the University of California Cooperative Extension (UCCE) on regional productivity in California agriculture. UCCE is responsible for agricultural research and development (R&D), and dissemination of agricultural know-how in the state.
Method/methodology/approach
We estimate the effect of UCCE on county-level agricultural productivity for the years 1992–2012, using an agricultural production function with measures of agricultural extension inputs alongside the traditional agricultural production inputs at the county level.
Findings
Results show a positive impact of UCCE through its stock of depreciated expenditures. For an additional dollar spent on UCCE expenditures stock, agricultural productivity, measured as value of sales at the county level, improves by $1–9 per acre of farmland for knowledge/expenditure depreciation rates between 0 and 20 percent.
Practical implications
Results suggest that county differences in productivity could affect extension expenditures. The high level of contribution found in the results would be especially useful during a period of political pressure to reduce public spending for agricultural extension in the state.
Theoretical implications
Theoretical implications suggest that agricultural systems with higher level of knowledge depreciation are associated with higher resulting incremental agricultural productivity per an additional dollar spent on UCCE expenditures stock. This suggests that extension policy should consider also the agricultural system (crop mix).
Originality
We use original budgetary data that was collected especially for answering our research questions from archives of UCCE. We estimate impact of extension at the county level in California, on the value of agricultural sales (of crops and livestock). We developed an extension expenditure stock, using current and past expenditures data, and different depreciation rates, following the theory of Knowledge Production Function.
Disclosure statement
No potential conflict of interest was reported by the authors.
Notes on contributors
Diti Chatterjee is an Economist with A2F Consulting in Bethesda, MD, USA. She works on aspects of economic investment and growth and environmental consequences in developing countries.
Ariel Dinar is a Distinguished Professor of Environmental Economics and Policy at the School of Public Policy, University of California, Riverside, USA. His research addresses agricultural economics and natural and environmental resource economics as well as economic aspects of agricultural extension.
Gloria González-Rivera is Professor of Economics at the University of California Riverside. She is the President and elected director to the board of the International Institute of Forecasters. Her research focuses on the development of econometric and forecasting methodology with applications in financial markets, volatility forecasting, risk management, and agricultural markets.
Notes
3 We use value of agricultural sales to measure agricultural productivity in this paper. Although these two terms are not the same because we do not account for cost of production. Value of agricultural sales is used in this analysis as a proxy of the county's agricultural sector productivity, following a similar methodology by OECD (Citation2001).
4 We use the concept of knowledge production function (KPF). The development of KPF is not the focus of this paper. We provide general information about our estimates, but the reader is referred for more details to Chatterjee, Dinar, and González-Rivera (Citation2018).
7 We are unable to account for undocumented labor in this analysis because the data on undocumented labor is not available in the USDA agricultural census. This data source does not provide information on family labor. However, Martin (Citation2018) suggests that the share of hired labor of the total labor employed in California is 65 percent during the year of the analysis in this paper. While we capture the majority of the labor employed on farms in California, it is likely that our results provide an overestimation of the impact of labor on value of agricultural sales per acre.
8 While production decisions are made at the farm level, this approach captures the results of such decisions and decisions of extension expenditure allocation on a per acre of farmland to control for size effects.
9 We use a simple linear relationship in this paper; other cases with non-linear relationships between the inputs and the dependent variable could be potentially used for the analysis. We estimated different models and decided to report the linear model coefficients. We do not consider the extreme cases in which there is only one input, UCCE expenditures, or labor, in this analysis.
10 The choice of the number of lags is also guided by unavailability of data beyond five lags.
11 For clarification, certain type of machinery could be more important than other types of machinery, and the impact of older machinery may have depreciated. The AG Census does ask respondents to indicate how old the machines are, and then specifically asks how many were used in the cultivation process that year. However, there is no way to know from the responses the exact type of machinery (and age) used by the farm. For that reason, we did not use the detailed information but instead used a count variable for machinery.
12 In the context of measuring production input at the county level, Huffman (Citation1976) uses different approaches for some of the inputs in the production process with data for 1960. For example, Huffman (Citation1976) uses family labor+hired labor, while we use only hired labor. Our data source (FRIS) provides only hired labor data for 2014, mainly because in 2014 the structure of the agricultural farms in California transformed, compared with 1960, and consist of much more hired labor (Martin Citation2018). While Huffman (Citation1976) had access to fertilizer input data in the form of price-weighted primary plant nutrients, that work doesn't include data on chemicals such as pesticides.
13 This is the mean value for UCCE expenditures per acre for each of the five census years.
14 The mean value of the share of acres of harvested land to total farmland acres calculated, based on our entire data set, equals 0.25. The calculated percentage of total acres harvested, to total acres of farmland (across all counties, and all years) amounts to 30 percent. This figure is very similar to that reported in the 2002 report by the University of California, Davis: http://aic.ucdavis.edu/publications/moca/moca_current/moca09/moca09chapter1.pdf. It is the result of increased water scarcity during the years for which we use FRIS data, leading to reduction in irrigated acres.
15 The variable that represents ‘chemicals and fertilizers’ is measured as the ratio of total area on which fertilizers, pesticides, and other chemicals were applied, to total area of farm land. In the Census of Agriculture, farmers are asked to provide a count of the number of acres on which four main types of chemicals are applied to treat diseases and two types of fertilizers are added, including manure. We create a count variable that is divided by total farmland acreage, and the resulting variable can theoretically range, for each farm surveyed, between 0 and n (n > 1). The reason is that the same acreage could be reported several times as receiving chemicals and fertilizers.
16 Census of Agriculture highlights, 2012. https://www.agcensus.usda.gov/Publications/2012/Preliminary_Report/Highlights.pdf
17 This is obtained by setting δ = 0 in Equations (4) and (5).
18 We estimated the same empirical model, including individual UCCE expenditure lags as separate independent variables. The estimates indicate that each individual expenditure lag does not have a statistically significant impact. The idea is similar to what the literature suggests. The underlying principle is that the expenditures stock, which generates a knowledge stock, affects productivity, or value of agricultural sales.
19 Expenditure on hired labor is obtained from the agricultural census reports published by USDA. It is divided by total number of hired labor recorded in the census, and then expressed in per acre terms through division by total farmland in acres, all values aggregated at the county level.
20 Expenditure on all chemical and fertilizer application is obtained from the agricultural census reports. It is divided by total number of acres on which application took place, and then expressed in per acre farmland terms through division by total farmland (acres); all values are aggregated at the county level.
21 The above model was estimated, including number of primary occupation farmers per farm for a county as the independent variable instead of number of primary occupations per county. This is to capture the cases in which a primary occupation farmer is cultivating more land and producing less output, or vice versa. The coefficient estimate of the new independent variable is still negative but statistically insignificant.
22 We have estimated a model, including county average temperature and precipitation into the regression model represented by Equation (3), and found that weather variables do not have any significant impact. We also estimated the model with interaction terms between UCCE expenditures stock and our county average temperature and precipitation, and obtained insignificant coefficients.
23 Twenty-seven counties with statistically insignificant coefficients were removed from the analysis to minimize the loss of degrees of freedom.
24 This is done due to space constraint.
25 Numbers are rounded.
26 These counties are all low-ranking counties, in terms of production value. Some of the counties had experienced reduction in agricultural land (https://www.cdfa.ca.gov/statistics/pdfs/2013/finaldraft2012-2013.pdf). These counties are located in mountainous regions and specialize in agricultural crops facing harsh market conditions (e.g. pasture) and have difficulties transforming UCCE knowledge into sales.
27 The coefficient estimate for UCCE is statistically insignificant in case of Los Angeles, and small but positive and statistically significant for San Francisco-San Mateo. Through discussions with UCCE officials, we learned that both of these counties include considerable non-agricultural research and outreach work done by UCCE that is not included in our analysis. This may explain why agricultural expenditures on research and outreach may not have any notable impact on agricultural sales in these two counties.
28 We were advised by an anonymous reviewer to try more consistent econometric models. We used procedures in Rabe-Hesketh and Skrondal (Citation2012) to estimate the random coefficient model and a model with consistent estimation of effects of endogeneous time-varying covariates (STATA commands xtmixed and xthtaylor). The results of these estimations yielded identical coefficients and significance levels to those obtained in our reported results. Therefore, we decided not to present the additional (identical) results. They can be provided upon request by the corresponding author.
29 This is calculated for knowledge depreciation rates ranging from 0 to 20 percent.
30 Few outliers were removed from the diagrams (only) to improve visibility of the names of counties.