We collected UCCE expenditure budget data by county offices for the years 1992 through 2012. Some counties in California have a shared budget allocation with another county; such counties include Humboldt and Del Norte, Inyo and Mono, Placer and Nevada, Plumas and Sierra, Sutter and Yuba, Shasta and Trinity, and San Mateo and San Francisco. We consider each of these two-county combinations as a single county for our analysis. There is no UCCE office in the county of Alpine in our records, so it is excluded from the analysis. We use the entire set of 50 counties, and we refer to UCCE budget data as extension expenditures. For our empirical model, ‘Land’ is measured as total harvested acres. ‘Labor’ is represented by total number of hired labor employed. ‘Machinery’ is the sum of all kinds of machines used in the production process for each county, which includes cotton pickers and strippers, forage harvesters, grain and bean combines, hay balers, tractors, and trucks, including pickups.The variable ‘chemicals’ is the ratio of the sum of all acreage on which fertilizers and pesticides were applied, to total farmland. We were constrained by the data in the Farm and Ranch Irrigation Survey regarding several inputs. Use of chemicals was one of them. FRIS reported the acres treated with chemicals, and this was our only source for input of chemicals. A higher share of farmland acreage that is treated with chemicals is expected to be associated with higher yields and thus higher sale values.We use the variable ‘average farmer age’ to represent farmer experience in a county. We also include total number of farmers in a county with farming as their primary occupation as a second farmer characteristic variable. Summary statistics for the entire data set are reported in Table 1. Mean total value of sales per acre forour data is $1,316, and UCCE expenditures per acre is nearly $6.One unit of hired labor is employed per 50 acres of farmland, and one unit of machinery is used per 100 acres. Chemicals and fertilizers are applied to nearly two-thirds of farmland acreage.The average farmer’s age in the state is 57.2 years.Figure 1 shows the relationship between total value of sales per acre and the stock of extension expenditures per acre for each year for all counties. We observe a positive correlation of 0.81, 0.66, 0.70, 0.53, and 0.25 between extension expenditures and productivity for the years 1992, 1997, 2002, 2007, and 2012, respectively. We also observe very high levels of extension expenditures for counties such as Los Angeles and San FranciscoSan Mateo.

Los Angeles county agricultural products include alfalfa, one of the most important crops in that region. Its average total value of sales per acre is $2,547, which is nearly double the sample mean of $1,316. Mean extension expenditures per acre for the county equals $30, fodder systems for cattle which is nearly five times the sample mean of $6.2. San Francisco–San Mateo counties include wine and apiary products as their most prominent crops. These two counties have the highest mean extension expenditures per acre of $45, and a mean total value of sales per acre of $3,283. Santa Cruz has the highest average value of sales per acre among all counties, which equals $6,902, and its mean extension expenditure per acre is $24. The highest amount of cash receipts for the county comes from strawberries, raspberries, and other berries, followed by nursery crops and vegetables like Brussels sprouts and lettuce. Mariposa has the lowest average value of sales per acre at $53, and $2 worth of average UCCE expenditures per acre. It can be argued that higher expenditures on research and extension in some of the lower-performing counties can be substitutes for other traditional inputs, which may be scarce in supply. With the availability of efficient methods of agriculture, higher income for farmers as well as lower-priced, home-grown crop production can be ensured for the county, thereby benefitting both the consumers and producers of agricultural products. Section 5.3 of the paper discusses the issue of substitutability of traditional inputs with extension expenditures for policy purposes.The annual UCCE budget data includes different expenditures, such as salaries, infrastructure, dissemination, and experiments cost. We roughly aggregate the expenditures into two groups: salaries and infrastructure . UCCE total annual expenditures range between $77 and $90 million . State-level expenditures by UCCE trended lower in the 1990s as can be seen in Figure 2, Panel . This trend was broken in the early and mid-2000s but, around the 2008 financial crisis, UCCE expenditures experienced further decline to reach the lowest level since 1992. In Figure 2, Panel , we can see that the total value of sales per acre of farmland in California has been growing over the five census years included in our analysis. Between the period of 1992 and 2012, the total value of agricultural sales per acre of farmland has risen by 90percent. Real expenditures made by UCCE normalized per acre have remained relatively unchanged over these census years, ranging between $3.1 and $3.6. Over the same period, we observe a 12 percent reduction in acres of farmland in the state, declining from 29 million acres in 1992 to 25.5 million in 2012. When we consider the stock of extension expenditures instead of the extension expenditures for the current year only, then the picture is different. In Figure 2 Panel , we observe that the stock of extension expenditures per acre has remained more or less stable around $20 per acre in 1997 and 2002, and has risen slightly in 2007 and 2012.

The value for 1992 is $3.4, which is about 20 percent of the mean for the rest of the period. The reason for this difference is the fact that we did not have data for UCCE expenditures in the five years prior to 1992. For the year 1992, aggregate state level data for five lagged values of UCCE expenditures were available. Upon computing the expenditure stock at the state level, using Equation , we observed that expenditures stock remained comparable to the four other census years. We used this result to assume that UCCE expenditures stock for each county remained comparable across the 5 census years. Making use of this assumption, we have inflated UCCE expenditure of 1992 for each county by a multiple of 5 in our empirical analyses. Our general observation is that even though the annual UCCE expenditures have fallen over time, the extension expenditure stock per acre of farmland has risen slightly over this period. We observe a decline in the extension expenditure stock in 2012; this may be the reflection of the effect of the steady decline in annual UCCE expenditures since 2009 that we observe in Figure 2 Panel . We observe also in Figure 2 Panel , a positive relationship between this cumulative input and sales per acre over the period of our study. Some of the funding for extension activities is statewide funding that should be included in the analysis. However, our experience with statewide funding of extension activities is such that we decided not to include it in our analysis. State funding for extension activities is adhoc and is the result of catastrophic events, such as flooding, diseases, and other local or regional disasters. Therefore, such funding is hard to find in one source, and its attributes could be different from year to year. We decided not to use this type of funding in the analysis. Our decision could lead to over-estimation of the contribution of budget dollars that are provided for extension.Empirical results for the county-level panel data analysis from Section 3 are reported in Table 2. We have considered a number of cases. In the first case, we considered knowledge depreciation rate to be zero. This implies that all old knowledge remains effective, and each of the five expenditure lags in the expenditure stock variable has equal impact. The coefficients for this regression are reported in column of Table 2. Results indicate that the coefficient for stock of UCCE extension expenditure equals 1.05, and is statistically different from zero at a 10 percent level of significance. This implies that a $1 increase in the expenditure stock leads to an extra $1.05 in the value of sales per acre, on average.

Harvested acres measured as a share of total farmland has a negative coefficient with total value of sales per acre,fodder sprouting system but this effect is not statistically different from zero.The marginal value of hired labor per acre of farmland is $23,262. Hired labor accounts for nearly 33 percent of all farm employment and is responsible for about 60 percent of all farm work in the U.S., according to Martin and Jackson-Smith . The labor force is largely born abroad and has become more important for larger farms in the country. Hired labor employment per acre has undergone a 22 percent decrease between 1992 and 2012 . In the case of California, hired labor constituted 65 percent of the farm workforce in 2014 .The average cost of hired labor is about $10,385 for our sample , with an average per acre cost of less than $1. Therefore, for additional hired labor, there is a net gain of nearly $23,000 in total value of sales per acre.Machinery has a positive coefficient, according to our findings, but the effect is not statistically significant. Acres on which chemicals were applied as a share of total acres of farmland show a statistically significant increase of $1,248 on county productivity. Our data indicate a 43 percent increase in acreage of chemical application as a share of total farmland over the period 1992–2012, contributing to the increase in productivity seen over the same period. Average cost of chemicals per acre of application is $152. When calculated in per acre farmland terms, this amount becomes less than $1.Therefore the $1,248 addition to the total value of sales per acre is also the net impact of an acre of chemical application per acre of farmland. An additional primary-occupation farmer in a county impacts productivity negatively in our analysis. This impact is statistically different from 0 at 1 percent level of significance, which could be interpreted as less-efficient producers in the agricultural sector whose primary occupation is farming. Previous literature provides empirical evidence of movement of educated and more efficient farmers to off-farm work, both for the U.S. and internationally for Pakistan . The more efficient farmers may have obtained multiple jobs or careers, thereby leaving the less efficient ones as primary-occupation farmers, which is captured by our estimated coefficient in Table 2. 21 Columns – of Table 2 report coefficient estimates for our original model with different values of knowledge depreciation rates .

For δ ranging between 5 and 9 percent, represented by columns, the coefficient of the expenditure stock variable changes from $6.70 to $7.18 at a 5 percent level of significance. The coefficients of other control variables are very similar for the different depreciation rates. The coefficient of the UCCE extension expenditure stock variable increases from $7.30 to $8.60, between 10 and 20 percent knowledge depreciation rate values. We test knowledge depreciation rates beyond 20 percent, which are reported in Appendix Table A1. Results indicate up to 49 percent depreciation of knowledge for our data set, and we observed positive, increasing coefficient estimates for UCCE extension expenditures stock. Beyond 50 percent, the coefficient estimate becomes statistically insignificant. This implies that the contribution of UCCE expenditure stock towards productivity improves under the assumption of a dynamic system in which old knowledge is replaced up to a threshold depreciation rate, while controlling for everything else. Beyond this rate of depreciation, the impact of UCCE extension expenditures stock on agricultural sales per acre becomes insignificant. Knowledge stock with a 100 percent knowledge depreciation rate is represented by current period UCCE extension expenditures. In such case, all previous expenditures become obsolete in terms of their effect. Regression results are reported in column of Table 2. We see that while coefficients for all other control variables remain similar to the depreciated knowledge cases, the coefficient for the extension expenditure stock variable becomes negative and significantly different from 0, at a 5 percent level of significance. This implies that current expenditures reduce current total value of sales per acre by nearly $31.