November 19, 2012

Using IVPROBIT: Do Farm Size and Farmer Attitudes Towards Planting Genetically-Modified Corn or Soybean Crops?

This is the question posed to us by Dr. Harvey James as an exercise. Using a survey of 3,000 agricultural producers that he and Dr. Mary Hendrickson did in 2006, we tested how farm size and attitudes toward genetically-modified organisms (GMO) affect a farmer's decision to plant GMO corn and/or soybean.

We limit the data to those that only produced corn and/or soybean crops because we want to focus only at those farmers. The main variables of interest are if the farmer has experienced planting a GMO crop (ADOPT) for the dependent variable, and log of farm size in acres (LNACRES) as well as positive attitude towards GMOs (ATTITUDE) as the main regressors. By the way, as a side note, it's always a good rule of thumb to use the log of a variable if the values of the variable are too large.

We also consider other control variables, such as age of the farmer (AGEGROUP), if the farmer also raises livestock (LIVESTOCK), and if the farmer attends church very often (CHURCH). It is expected that older farmers are less likely to adopt (rely more on traditional farming), that farmers who raise livestock are less likely to adopt, and that farmers who attend church more often are less likely to adopt (probably because more religous farmers view genetic manipulation negatively).

Automatically, we use binary choice models (probit in this case) as the econometric method. The results are as follows:



Well, except for CHURCH, we have the expected results. Older farmers and those that raise livestock are less likely to adopt GMOs. The more important results, however, are the first two regressors: the larger the farm, the more likely the farmer adopts; and (ho-hum) if a farmer has positive attitude towards GMOs, then he or she is more likely to plant GMOs.

Now here's the catch: there might be an endogeneity problem. If you are already planting GMOs, would you say that GMOs are not good for farmers? The answer is no. The fact that you are already adopting GMOs, the more likely you have a positive attitude toward GMOs.

So, for endogeneity problems, we use instrumental variables approach. It is more common to have instrumental variable regressions if the dependent variable is continuous.

What about binary variables such as ADOPT, which is a simple yes if you planted GMOs and no otherwise? Good thing there is also the instrumental variables probit model, which can easily be implemented in STATA with the IVPROB command.

Now, it is given that farmers who are optimistic about the future are likely to have a positive attitude towards GMO adoption, but that this optimism is totally unrelated to farmers' decisions of adopting GMOs or not. And so we use this OPTIMISM as the instrumental variable.

Now like all instrumental variable estimation, there are actually two methods: two-stage least squares estimation (using the TWOSTEP option in STATA) and the simultaneous maximum likelihood estimation (using the MLE option in STATA). We preferred the simplified, reduced-form approach and so we went for the two-stage method. The results are as follows:



Well, surprise surprise. The signs are still there, but only two remain as significant: size of farm and livestock. Age and attitude doesn't affect GMO adoption? We will have to take that with suspicion. And we may be correct. Because if you look at the Wald test for exogeneity (at the bottom of the table), the statistic is insignificant. This only means one thing: THERE IS NO ENDOGENEITY. Well, at least none for this sample. Remember, we just said there MAY be endogeneity. But looking at these results, there is actually none.

So given that, we can go back and report the probit result as the main one. Corn and/or soybean producers that have larger farms are more likely to adopt GM crops. Those producers that have positive attitudes towards GMOs are also more likely to adopt GM crops.

You'll have to put an additional "DUH" on that second one.

May 3, 2012

Put a Little Shade in your Graphs

Suppose you graph a time series and you want to highlight a certain range (for example, U.S. real GDP series and you want to color the area between 1929 to 1939 to indicate the years under the Great Depression):

line rgdp year

Well, a simple way to do that is to add another variable and use the twoway graph command "AREA." To start with, your new variable has to be a constant with value equal to the highest data point in your y-axis. So for example, in your graph of U.S. real GDP, the highest value given is 3,500 billions of constant 2005 $. So:

gen new=3500

Now, your new graph command should be:

twoway (area new year if year>=1929 & year<=1939) (line rgdp year)

You have to start with the AREA-type graph first because the first graph type will always be superimposed by the next graph type.

Special Characters in STATA Graphs

Thanks to helpful hints from my friend Mitch Abdon's excellent STATA blog site, if you want to use special characters in your STATA graphs (e.g. Greek letters), you can look at full options by typing:

help graph_text