1

Is fixed-effects regression invalid if you don't have panel data? My study is about the impact of microfinance loans in poverty alleviation.

gung - Reinstate Monica
  • 145,122
Gonzalo Ruales
  • 11
  • 1
    You will get no good answers to this question without providing more detail. What you do you mean by "invalid?" In what would would regression models be "invalid" without panel data? Regression modelling was initially created for non-panel data, so I'm not sure why you would think it would be invalid if you lacked panel data. Can you clarify your question more? In what manner were you expending a longitudinal data component? – StatsStudent Oct 22 '15 at 03:25
  • 1
    Now that clarifies something.. not really a stat expert here.. so, you mean a regression model is designed for non-panel data.. I am going to use a fixed-effect regression for an impact study wherein the conduct would be a one-time survey.. Most impact studies I have read require time-series data which I cannot afford to gather within a short span of time,, Thanks much.. – Gonzalo Ruales Oct 22 '15 at 06:18
  • In a sense, there are many different types of "regression" models that can be fit to data. Regression was used on data that was not panel-data, but over time, advances in statistical methods have seen the development of regression methods that now can accommodate panel data. So, regression can be used for either panel data or non-panel data. It sounds like your study will be a fine fit for regular ordinary least squares regression analysis, depending on the type of dependent variable you have. If your dependent variable is not continuous, you may need to use different regression models. – StatsStudent Oct 22 '15 at 16:39

1 Answers1

1

You can run a fixed effects regression on such data if you have a grouping variable to which multiple observations belong. Examples of such a variable in your case might be lender or locality. You need a decent number of observations per grouping for fixed effects to be helpful (say 7-10), but provided you do, you can include fixed effects for your grouping variable. All it does is estimate effects of interest within groups, which may or may not be useful.

Edit. The last statement is the idea that fixed effects estimates are not meant to generalize to the population, but are purely for making inferences about your sample.

Erik Ruzek
  • 4,640
  • I really appreciate it. Does it mean I could have a one-time survey? One of my dependent/outcome variables is per capita income. My factors/independent/treatment variables are age, number of years in the village, no. of cycles, etc. Also, can I use SPSS to run my data? It's the only software available in our school. Thanks much.. – Gonzalo Ruales Oct 22 '15 at 03:40
  • The factors/independent variables you list do not seem to fit a typical problem that fixed effects addresses. That is, fixed effects helps you when a gouping variable (typically at a higher level of your data hierarchy) may have undue influence on the effect of your focal independent variable, and you only want to make comparisons among observations within that grouping variable. As for whether you can do this in SPSS, I don't know for sure, but I'd be shocked if there wasn't a procedure for this. Worst case scenario is to use a set of dummy variables for your grouping variable. – Erik Ruzek Oct 23 '15 at 00:33