Creating a variable for positive and for negative change

[Leo Nuss]

Dear Stataticians,

I am right now developing my first empirical model for a research paper. I want to measure if the quality of political system in Russia is responsive to energy sanctions.

In my model I understand energy sanctions as negative changes in Russian oil and gas revenues and ask whether the political system responds significantly to positive and negative changes in oil and gas revenues by looking at historical data.

I want to use am OLS regression analysis using time series methods in which the dependent variable is the Polity index.
Now I shall use (my professor says) as my main independent variables positive and negative changes in oil and gas export revenues (per capita) in addition to a couple of other drivers of Polity.

Hence, I should create two variables one for positive and the other for negative changes of oil and gas revenues. However, I have no idea how to do that. I have data on fuel exports as percentage of merchandise exports.
Can I do anything with that?

Thank you in advance!

[Clyde Schechter]

Well, there are two different stages here. On the one hand you have data on fuel exports as a percentage of merchandise exports, but you need to get oil and gas revenues. So that will involve come kind of calculations, probably involving additional data beyond what you mention here, and this would be a question of understanding accounting and economics, not a statistical or Stata issue.

The second stage concerns deriving the positive and negative changes from the actual revenue. So the first part is calculating the actual change:

tsset year
gen revenue_change = D1.revenue

Then there is the question of representing positive and negative changes. If you really want to create two separate variables, one denoting positive change, and the other denoting negative, the code for that would be:

gen positive_change = (revenue_change > 0)
gen negative_change = (revenue_change < 0)

But that is probably not a useful thing to do. First of all, except for the unlikely possibility of an exactly zero change in revenue, these two variables are just opposites of each other, so you really only need one of them. If the change isn't positive, it's negative, and vice versa (again, unless it is zero, which rarely happens.)

But let me challenge you more. Making a dichotomy out of a continuous variable is usually a bad idea: it throws away information. When people model performance of political parties in elections, they usually do not use win-loss as their variable, they model the point spread. Similarly, when sports teams are studied for their performance, win-loss is not the variable of interest, point spread is.

So my guess is that you will have a better model if you forget about creating a variable for positive or negative change, and just use the actual revenue change itself in your analysis. If negative revenue shocks are associated with an increased polity index, you will see that as a negative regression coefficient for the revenue change variable.