This negative coefficient could capture the allocation of a higher value of resources for counties that have experienced low performance during that fiscal year, or cutbacks for a particular county that has performed well. As Foster and Rosenzweig found new technology takes a while to be adopted, and its full impact is observed over time. So, a combination of the two may explain the results we obtained. Therefore, consideration of only the current period expenditures on measuring the impact of UCCE research and outreach expenditures on productivity only tells part of the story. A more complete picture requires understanding how the current stock of research-based knowledge impacts productivity. The current knowledge stock is the sum of old and new knowledge produced through expenditures on R&D and outreach, thereby providing a more complete understanding of the long-term impact of UCCE expenditure on county productivity.22 The trend we observe in the UCCE extension expenditure coefficients as the rate of depreciation grows from 0 to 100 percent presents an increase up to 50 percent depreciation and then a decline. At that range, we observe either insignificant coefficients or negative coefficients.One possible interpretation is that the more frequent the replacement of knowledge the higher the impact of funds spent on knowledge creation and dissemination. This is up to a given point at which the effectiveness of the knowledge stock decreases/drops. When knowledge replacement is 100 percent, meaning every year all knowledge becomes obsolete and needs to be replaced, the UCCE system is not efficient, leading to a negative coefficient of its expenditures stock.Empirical results in Table 2 inform how UCCE impacts average county-level productivity. However, we now want to test how the impact of UCCE expenditure on productivity varies across counties. Heterogeneous impact across counties can result from various reasons. In particular, differences in the resource base in the various counties,microgreen fodder system and the composition of the crops grown, can lead to differences in extension productivity.

From a policy perspective, this analysis is an important contribution to the literature, because it allows evaluating policies that affect certain localities that face different climatic or soil fertility. To achieve this, we have made some modifications to our original model. The main empirical model remains unchanged, but we include interaction terms between dummy variables representing each county and its UCCE expenditures into the old model. Regression coefficients for 23 counties are reported in Table 3 for knowledge depreciation rates ranging from 0 to 20 percent, and it includes only the estimates of the coefficients for the counties that interacted with the UCCE expenditures.The first row in Table 3 reports the impact of UCCE in Alameda County on total value of sales, which is negative for all used knowledge depreciation rates, and is statistically insignificant.Fresno County records the highest positive coefficient of UCCE expenditures stock. It varies from $25 to $191, depending on total value of sales per acre, for knowledge depreciation rates ranging between 0 and 20 percent, respectively. The coefficients for Fresno County are the highest and statistically different from 0 at 1 percent level of significance. San Bernardino County has the next highest impact on total value of sales per acre, which ranges between $10 and $82. Thethird highest statistically significant impact is obtained for Tulare County, which ranges between $10 and $72. The coefficient estimates for Los Angeles and Santa Clara counties indicate no significant impact of UCCE expenditures stock. Kern, Monterey, San Joaquin, Stanislaus, and Ventura counties, which are among the top 10 agricultural counties, have positive and statistically significant impacts reported in columns – of Table 3. Amador, Calaveras, Humboldt-Del Norte, Modoc, and Siskiyou counties have negative statistically significant coefficient estimates for knowledge depreciation rates ranging from 0 to 20 percent.For Imperial County, we observe that for 20 percent knowledge depreciation rate, the value of the coefficient estimate does not remain statistically different from 0. This result implies that adoption of new technologies at these rates may incur high costs and can stop impacting productivity positively. Los Angeles, San Francisco-San Mateo, and Santa Cruz counties do not report high impact on productivity, even though they are among the counties recording some of the highest expenditures made by UCCE.Overall, Fresno, Kern, Monterey, Tulare, and San Bernardino counties record the largest impacts of UCCE expenditure stock.

The first four counties are among the top 10 agricultural producers in the state. All these counties are also among the biggest producers of some of the most high-profile agricultural products in terms of receipts, e.g. grapes, almonds, strawberries, and citrus among fruits and nuts, tomatoes and lettuce among vegetables, and dairy, livestock, and poultry. The results discussed above provide better understanding of UCCE’s impact on individual county-level productivity. More productive counties in general report higher impact of UCCE presence.A pertinent issue with respect to this paper is the substitutability between UCCE expenditure stock and other inputs of production. This is particularly relevant because some counties may face scarcity of one or more of the traditional inputs, and it would be an important contribution if expenditures on UCCE can be a substitute for the said input. For this analysis, we use the inputs that have been found to have a statistically significant positive impact on productivity, such as hired labor, and acres of chemical application. Since number of primary occupation farmers brings down productivity, it is a ‘bad’ input. We have used a linear model in this paper, which makes the calculations simpler, under the assumption of constant marginal productivity. This means that a $1 increase in UCCE extension expenditure stock per acre of farmland will lead to a reduction in hired labor per acre by nearly 0.0003 workers, keeping total value of sales per acre constant. This is a reduction of nearly 1.5 percent, compared to the mean value of this variable . For the next significant input, which is acres of chemicals applied as a share of total farmland acres, we find that MRTS equals −0.00556 . This means that a $1 increase in UCCE expenditure stock per acre of farmland will lead to reduction in the share of chemicals applied per acre by nearly 0.006, keeping total value of sales per acre constant. This again is a reduction of about 1 percent, compared to the mean value of this variable . Similar trends in substitution were reported in Goodhue, Klonsky, and Mohapatra , suggesting that almond grower education programs can have a significant effect on pesticide use decisions.

We observe that substitution effect is low between the aforementioned traditional inputs and UCCE expenditures, thereby hinting at complementarity between each of them and UCCE expenditures. These estimates are a starting point in the discussion on the topic, which has very important policy implications not only for California but also for the entire nation. Using the coefficient estimates, we calculate the rise in total value of sales per acre for our sample, using mean UCCE extension expenditures per acre. That amounts to $41 . Multiplying this $ value by mean farmland acres in our dataset over the analyzed period provides a total increase in value of sales amounting to $22,165,359 , on average, per county. The average per county real UCCE extension expenditure for the 20-year period between 1992 and 2012 amounts to $1,778,146, which implies an average per county profit of nearly $20 million , due to the UCCE extension expenditures on research and development, and outreach. This provides some evidence of the scale of impact UCCE expenditures stock has on average county productivity. The same calculations for individual counties can provide a more in-depth understanding of the effects on them for policy planning.We observe allocated extension expenditures per acre with a mean of 6.21 and a standard deviation of 8.59, suggesting a wide difference across the counties. Decisions on allocation of extension funding at the county level depend on many criteria, including the county’s productivity and long-term planning criteria , and even political considerations and lobbying, as was already suggested in our paper. While we are running contemporaneous regressions , the decision making does not take place contemporaneously. That is, the yearly spending budgets are set prior to whatever economic activity goes on in the county during that year. The budget process happens before the total sales for that year are known. It seems natural to think that the effect runs from spending budget to sales,barley fodder system rather than the other way around. Nevertheless, we have only five years of production and sales data, and it may be reasonable to think that productive counties in a year can be favored with larger budgets the following years. But this is not the case. Scrutiny of Figure 1 suggests, for example, that Fresno, which is one of the most productive counties, had UCCE expenditure of about $3 per acre in 1992 and in 2007. On the other extreme, Alameda, which is one of the least productive counties, had UCCE expenditure of about $7 per acre in 1992 and in 2007. Thus, over time we do not observe overall big changes in UCCE expenditure that are triggered by the productivity of the county, and the case for endogeneity becomes weaker, if not irrelevant.

The large SD of the extension expenditures reinforces the aim of our analysis that explains the variation in sales as a function of the variability of UCCE expenditures. A caveat of this paper is that spillover effects across counties have not been included in the model. The empirical model assumes that there is no spillover, but this effect can be incorporated in future work. This paper estimates a simplified model of agricultural sales as a function of inputs, including UCCE expenditures stock, to provide a county-level impact of UCCE expenditures on R&D and outreach on productivity, which can provide policymakers with a reference point for policy decisions in California. Another caveat is the relatively short period of time , considered in our analysis. Longer time-series data would lead to higher values of benefits from the estimated impact equations.We estimate the impact of the University of California Cooperative Extension on county-level agricultural productivity in California, using a model representing a relationship between value of agricultural sales as a proxy for productivity, and quantitative inputs of production, including UCCE expenditures. Our analysis is aggregated to the county level because UCCE operates from county offices across the state. We obtained data for UCCE budgets for all agricultural research and development , and outreach/dissemination projects for 50 county offices statewide for the years 1992–2012 . Stock of knowledge produced through UCCE extension expenditures on R&D and outreach is modeled as a function of a stream of current and depreciated past expenditures, and used as our independent variable. Data on factors of agricultural production, such as harvested acreage, hired labor, chemical applications, machinery, average farmer age, and number of primary occupation farmers were obtained from the Census of Agriculture conducted by United States Department of Agriculture for five census years, spanning over 1992–2012. Productivity is represented by total value of sales per acre of farmland, using data from the Census of Agriculture. To estimate the impact of UCCE expenditures on agricultural R&D and outreach/dissemination on productivity, we construct a stock of expenditures. We use current and five lagged values of UCCE expenditures, and a range of different depreciation rates from the literature. The intuition is that old knowledge depreciates over time, therefore older expenditures enter the model at a depreciated value. We analyze our model using depreciation rates ranging from 0 to 9 percent, and then 10, 15, and 20 percent following Griliches . Regression results indicate that UCCE’s stock of expenditures has a statistically significant impact on total value of sales per acre, which varies from nearly $1 to $9, for depreciation rates between 0 and 20 percent. For higher rates of depreciation of expenditure, the coefficient becomes statistically insignificant. Results therefore suggest that for more dynamic systems with frequent innovations, UCCE’s efforts have a higher impact on productivity. This effect, however, becomes insignificant with very high levels of depreciation. For a knowledge depreciation rate of 100 percent, we find that the coefficient becomes negative , and this effect is statistically different from 0. This result likely captures the allocation of higher expenditures on counties that have reported lower performance during the year, or cutbacks for a particular county that is performing well.