Dear All, I have first-difference estimates using a two-period panel data-set with some of the variables being binary (e.g. whether or not households accessed credit). As Professor Jeff Wooldridge has been indicating (see, for example, e.g.http://www.statalist.org/forums/foru...ummy-variables)
there should be no problem differencing dummy variables. Nonetheless, I am still not sure how to interpret such differenced binary variables in terms of magnitude of effect. In my case, change in investment expenditure ∆ Inv (in US$) is the dependent variable. I suppose that one could say for a positive coefficient (on the differenced credit access variable, ∆ C) that investment growth increases with growth in access to credit. Am I right? Lets say that the coefficient on ∆ C is 0.698, what will be the interpretation in terms of magnitude of effect given that after differencing a binary variable takes on the values -1, 0, and 1?
there should be no problem differencing dummy variables. Nonetheless, I am still not sure how to interpret such differenced binary variables in terms of magnitude of effect. In my case, change in investment expenditure ∆ Inv (in US$) is the dependent variable. I suppose that one could say for a positive coefficient (on the differenced credit access variable, ∆ C) that investment growth increases with growth in access to credit. Am I right? Lets say that the coefficient on ∆ C is 0.698, what will be the interpretation in terms of magnitude of effect given that after differencing a binary variable takes on the values -1, 0, and 1?
Comment