The expenditures made by UCCE are shown in panel of Fig. 1. There is evidence of impact of the 2009–2010 recession on investments in 2010, which went down from over $90 million to less than $85 million. From 2010 onwards, we observe a steady decline in annual UCCE expenditures, to about $76 million in 2013. In 2007, the county offices that recorded some of the largest overall expenditures include Fresno, Tulare, San Diego, Humboldt-Del Norte, San Joaquin, Ventura, and Kern, in declining order. In 2013 we notice that leading counties in terms of overall expenditures were San Diego, Tulare, Kern, Plumas-Sierra, San Francisco, and San Mateo. Data on salaries of advisors employed in each county office was collected from the UCCE database as well. Expenditures on infrastructure are the amount remaining in the budget after subtracting total expenditures on salaries for the counties. These expenditures capture non-salary related expenditures, including benefits and travel provisions for county advisors, along with various expenditures on research and outreach programs taken up by the county offices. Full-time equivalent employment data was obtained for advisors employed by each county office. We observed an overall fall in both advisor FTE and advisor salaries, as represented in panel , Fig. 1. After 2010, both FTE and expenditures on salaries showed consistent decline. We observe , Fig. 1 an overall declining trend in expenditures on infrastructure, with a fall of about $5 million between 2009 and 2010.
This could be the effect of the 2009–2010 recession,dutch bucket hydroponic which also led to a fall in overall expenditures during that period. Panel reflects the decline in total expenditures that include both salaries and non-salary infrastructure related expenditures. The outcome variable in our empirical analysis is created using data on a number of component variables. UC ANR records data on a variety of methods in which knowledge, produced through investments in research and infrastructure, is disseminated. We use knowledge produced and knowledge disseminated interchangeably, because all knowledge produced by UCCE is publicly available and is disseminated. Hence, the methods of dissemination capture knowledge produced. These methods are categorized into three main knowledge groups. The first group includes data on classes, workshops, demonstrations, individual consultations, meetings or group discussions, educational presentations at meetings, and all other kinds of direct extension activities. The variable is named direct contact knowledge, and it includes all counts of knowledge dissemination from direct contact with growers. The second group is named indirect contact knowledge, and it includes counts of knowledge disseminated through indirect contact with possible clients via newsletters published and websites managed by UC ANR, television, radio programs or public service announcements, social marketing methods, mass-media efforts of knowledge dissemination, and other indirect extension efforts, including those through collaboration with other agencies. The last category is named research publication and other creative activity related knowledge.
This category includes counts of basic, applied or development research projects, program evaluation research projects, needs assessment research projects, educational products created via video and other digital media, curricula, and manuals created for educational purposes. We also include publications in peer-reviewed journals in this category. The above data on knowledge was recorded as counts. We were unable to categorize input variables into issues related to agriculture only, so to avoid overestimation issues, we include knowledge produced for all programs undertaken by UCCE for the period of the study. Using the data on all knowledge categories, we generated an index of knowledge as a weighted average of all the categories.6 We assigned weights to each category, based on relative importance of each kind of knowledge variable in terms of effectiveness. For this, we sent an electronic survey to the directors of all UCCE county offices in California. In the survey, we indicated the three above mentioned broad categories of knowledge production, with a number of subcategories. Respondents provided percentage weights for each broad category so the sum would add up to 100%. Within each broad category, respondents indicated percentage weights for each subcategory so the sum of the weights also equaled 100%. We obtained 10 replies from county directors after two rounds of surveys and created weights from the survey results. The completed surveys indicated that the most important effect on agricultural productivity is direct contact with farmers , followed by indirect contact with farmers , and finally research and publications . From the data collected on knowledge production variables, we identified seven federal planned programs : Climate Change, Healthy Families and Communities, Sustainable Food Systems, Water Quality, Quantity, and Security, Sustainable Energy, Endemic and Invasive Pests and Diseases, and Sustainable Natural Ecosystems. Climate Change was dropped from the official FPP categories from fiscal year 2013. Knowledge produced through indirect methods of contact is the most popular means of knowledge production, due to the comparatively lower cost of dissemination and wider reach to potential clientele.
Research projects, peer-reviewed publications, and the knowledge produced through them are also available to the public, but perhaps cater to a smaller audience compared to the other two methods. However, they are certainly a significant component in the direct interactions with farmers by specialists and county advisors. Over the period 2007–2013, we observe that all knowledge production declined as is illustrated in Fig. 2. Total knowledge produced in direct contact, indirect contact, and publication and research project methods of production have declined over time. Total number of counts of knowledge produced through all direct contact methods rose by 43%, from 15,059 in 2007 to 21,479 in 2011, but thereafter it continued falling until it reached a total count of 8282 in 2013, which is a 61% decrease compared to 2011. Knowledge produced through different methods of indirect contact with growers starts at 259,065 in 2007, and peaks at 405,386 in 2009, before falling down to nearly 43,000 counts per year in 2010. In 2013, the recorded number is 100,919, which is equivalent to a 61% reduction from the original levels in 2007. Research projects and peer-reviewed journal publications went down from 3349 in 2007 to 506 in 2013,dutch buckets system which is a percentage decline of nearly 85% of the 2007 value. Among all the counties, San Diego recorded the highest average count of knowledge production from direct methods, at 17147 , and Madera the lowest, at 3 . San Joaquin had the highest average count of knowledge production from indirect contact method at 49,225 , and Madera the lowest, at 0. San Luis Obispo had the highest value of average knowledge production through publications and research projects, at 308 , and Mariposa the lowest, at 1 . We also observe an overall falling trend in both inputs of knowledge production, such as county-level FTE, expenditures on salaries per unit FTE, expenditures on infrastructure per unit FTE, as well as output . In the next section, we report the results of our econometric estimates of the knowledge production function.Summary statistics of the variables in our analysis are reported in Table 1. We observe high levels of dispersion in the distribution of some of the knowledge variables. At the county level, San Joaquin, one of the most important agricultural producers, presents the highest mean knowledge index over 2007–2013, while Madera had the lowest. Mean advisor FTE number in San Joaquin was 353% higher than that in Madera; with 36% lower expenditures on salaries per unit FTE, and a 1% lower expenditures on infrastructure per FTE, compared to Madera county. The knowledge index, the weighted average of counts of the component variables, had been declining for the period of our study, as seen in Fig. 3. The cross-sectional average value of log went down from about 3.9 to about 2.75 over the period of 2007–2013, which reflected a 68% decline in the knowledge index. With these observations, it is important to know how our inputs impacted the average knowledge produced, and how these declining trends in inputs may have impacted knowledge production. Similar trends in knowledge production in agriculture are reported also by Alston et al. and by Ball et al. for the USA as a whole. Table 2 reports the regression results of Eq. , including two different models. Column reports the results for the case in which we include county and year level dummy variables to control for any factors that remain fixed across counties or years, possibly impacting the dependent variable.
This is a noticeable contribution to the literature because recent works on agricultural knowledge production function estimates have been focused on state level or national level. However, decisions on allocation of funding for knowledge production in extension activities have been made at the county level. The second version of the model includes a time trend instead of time-fixed effects. The specification with time trends allows to treat time effects on knowledge production as a continuous rather than fixed effect variable, which potentially can be more useful for policy makers. In the case of our analysis, these two models produced very similar results as is discussed below. We obtained statistically significant coefficients for all the input variables in both versions of our model reported in columns Model and Model of Table 2. A percentage rise in FTE impacted knowledge production positively by nearly 1.1%. A 1% rise in expenditures on salaries per unit FTE increased knowledge production by 0.86%. The coefficient estimate for the linear term of expenditures on infrastructures per unit FTE is positive and the coefficient estimate for the quadratic term is negative, supporting the theory of diminishing marginal returns to expenditures in infrastructure per FTE employee. In Model , we controlled for county-level fixed effects by introducing county dummy variables. Here, we de-trended the dependent variable as well as the independent variables by including a time trend variable in the model. We reported robust standard errors in the parentheses.Coefficient estimates for both the models are comparable to each other. While it is difficult to compare our results in Table 2 for an agricultural research and extension system to results of work on industrial knowledge production function, still there are several similarities in terms of the relative importance and the sign of the coefficients of the estimated knowledge function to the work of Czarnitzki et al. . We computed the elasticities of production, based on results in Table 2, which are reported in Table 3 below. The elasticity of production of knowledge with respect to FTE varied from 1.07 and 1.10, across the two models we estimated. The elasticity of knowledge production with respect to salary level varied between 0.86 and 0.87 across the two estimated models. The elasticity of knowledge production with respect to infrastructure expenditures varied between −0.39 and −0.31 across the two estimated models. The interpretation of these estimates is as follows: A 1% increase in FTE led to a 1.1% increase in average knowledge produced. Similarly, a 1% increase in expenditures on salaries per unit FTE would bring about a 0.87% increment in average knowledge produced by UCCE. The elasticity for expenditures on infrastructures per FTE for both models were calculated at the sample mean of this variable , using Eq. , as reported in Table 3. This value is negative, both in Model , and Model . Due to diminishing marginal returns, the relationship between this input and knowledge produced is concave, and the elasticity therefore depends upon the value of expenditures at which it is calculated. We computed the value of expenditures on infrastructure per unit FTE that corresponds to the turning point of the production function from a positive to a negative slope; this value equals $312,320.Expenditures on infrastructure per FTE less than this amount will yield a positive output elasticity; higher values will yield negative output elasticity, as is the case when we use the mean value. We observed that FTE is the most effective input in the knowledge production process, with an elasticity > 1. The advisor FTE employed by the county offices are engaged in various kinds of research and outreach operations and are the most important factor in the process of knowledge production. Dinar found similar evidence of significant positive marginal product of senior researchers on production of knowledge for the public agricultural research system in Israel.