I did efficiency analysis using Data Envelopment Analysis (DEA) and I am using these efficiency scores as a dependent variable. My independent variables are farmers' socioeconomics (age, education, farming experience, family member is farming, area under cotton) and institutional variables (access to extension, credit and market).
I have applied truncated regression, censored regression and Tobit model. But each of these models does not give many significant variables and most of the signs are negative too, except education. So, where am I doing wrong? Is the model selection not appropriate or there isn't necessarily a relationship among these variables? I found in literature, many researchers have found significant relationships using these models.
Any useful comments will be highly appreciated.
I do not know what is efficiency score. Perhaps some index derived on public opinion or other subjective variable. I guess that you need more objective variables, like output per unit of land, both physical (tons per ha) and revenue. Note that climate and soil can also influence output of a farm.
In a follow-up regression analysis of DEA efficiency scores, important are the choice of both (i) the regression method, and (ii) the explanatory variables. In econometric models the omission of important explanatory variables can lead to biased estimators and to invalidation of inference procedures.
In addition to socio-economic (i.e. farmer-specific) characteristics likely relevant are farm and field characteristics and also the specific technology/seed type used to grow the crop. It would be important to include yield potential given the soil type and weather conditions. Where this information is not available an alternative would be to include binary variables (dummies) to distinguish by sub-region. Dummies could also be used to differentiate farms by technology/seed type.
What's about the model fit? If the model is showing good fit, then there is possible multicollinearity.
In addition to the suggestions given above by the colleagues, please check your sample size too
What are the inputs and outputs for your DEA analysis? What relationship do these have to the explanatory variables?
Farhard. As I understand, you are first deriving efficiency scores using (some conceptual input-output model and) DEA method and then you use those scores as dependent variable in some other model in order to assess determinants of efficiency. This is known to be a two-stage approach. Wang and Schmidt (2002) have shown that such two-stage approaches to analyzing determinants of efficiency differences are seriously biased. See their article for alternatives (one-stage approaches). You can also see Battese and Coelli (1995) or, a more advanced approach by Koop, Osiewalski and Steel (2000; Journal of Business & Economic Statistics). It does not mean that everything is ok with your choice of variables and/or number of observations you have (DEA is a nonparametric method, thus results are more depended on data quality and sample size, which is often too small in empirical applications of DEA) but without addressing the "two-stage" problem you are unlikely to get anything valid out of this.
Hope this helps.
First you find the nature of relationship between the individual socioeconomic variables vis-a-vis technical efficiency scores using scatter plot. If there is linear/ non-linear relationship, accordingly fit your regression model for better estimates
Thank you all for your valuable suggestions and comments. I thought it may be appropriate to share that I have finalized it with improvements from your suggestions. Here is what I have got now.
After revising the indicators, I was particularly worried about the direction of one of the independent variable results: the adoption of "better cotton" practices. As suggested, I updated the explanatory variables but found that beside other variables the higher adoption of these improved management practices lead to lower economic efficiency. A deeper look into the data and field observations revealed that these practices do not increase yield immediately but are beneficial in the long run. So, the current negative relationship made sense.
@Yuri Yegorov: The efficiency scores are DEA efficiency scores whereby dependent variable was cotton yield (kg/ha) and independent variables were see quantity (kg/ha), irrigation (no./crop season), fertilizer (kg/ha), biofertilizer (kg/ha) and pesticides (no./crop season).
@David Tripe, the above variables were output and inputs.
Why should socio-economic factors impact on any of these variables? Would there be a theory to support any relationship, or would it just be chance?
@David Tripe: Actually there is extensive literature available which supports this kind of analysis whereby the efficiency scores (having similar output as inputs as mine) are regressed through OLS, Tobit or Truncated regression with socioeconomic variables as independents.
Theoretically, the efficiency of a farm is indeed related to these variables. For example, the education level of a farmer affects the farm efficiency because the farmer with higher education better understands the input-output relationship (which is measured by technical efficiency). Thus, these socioeconomic variables should be related to efficiency.
Stochastic Frontier Analysis (SFA) and Data Envelopment Analysis (DEA) are useful to estimate efficiencies over time. Stochastic frontier models allow to analyse technical inefficiency in the concept of production functions. Inefficiencies can be due to structural problems or market imperfections and other factors.
Perhaps, You can apply the stochastic frontier analysis.
Recently saw a paper presentation by one of our faculty member, he got significant relationship between profit efficiency (SFA) and socioeconomic varriables of farmers in China.
- Badges
- Science topic
- Similar topics
- Econ
- Manufacturing
- Industry