Empirical research can have no theoretical model; But if there is a good theoretical model as the basis, it will be more convincing.

Regression analysis can only explain the correlation between variables, and often needs economic theory to judge the causal relationship between variables.

Even if we can't provide a complete theoretical model, we should make some theoretical analysis.

Ideally, the econometric model is derived from the theoretical model, that is, the regression equation to be estimated.

A good empirical paper needs to tell a good "story" and then use data to confirm or test the story.

For the specific form of regression function, we can consider linear, logarithmic (variables only take positive numbers and have an exponential growth trend), double logarithmic, nonlinear (marginal effect is not constant) and so on.

When setting the model, we should try our best to use common sense and economic theory. For example, match "per capita variable" (such as per capita consumption) with "per capita variable" (such as per capita GDP); Explain the actual import and export with the real exchange rate.

For another example, consider the role of foreign direct investment in economic growth. Because it takes time for FDI to take effect, it may not be meaningful to restore the growth rate of that year to FDI of that year. It is more appropriate to consider the effect of FDI at the beginning on the economic growth in the following five years (or years) (which can also alleviate the two-way causality).

If you are not sure how to set up the measurement model, you can learn from the model setting of similar studies in the literature.

The model should be neither too simple (too few explanatory variables) nor too complicated, but it should be kept reasonably simple.

When choosing explanatory variables, the modeling method of "from concrete to general" is simple and easy, but there may be a big deviation (because there are missing variables);

The modeling method of "general to concrete" has small deviation, but it is not easy to realize. In practice, a compromise scheme is often adopted, that is, a simple and clear model is chosen.