Econometrics
There are many common econometric mistakes made in proposals submitted. I will provide a list of these mistakes, and how to recognize them, and how to fix them here. Most textbooks do not provide adequate guidance about these mistakes, which is why they are very common.
Common Econometric Errors:
The goal of research is to estimate a regression model.
Every question of the type: "what is the effect of X1, X2, X3 on Y?" makes sense and can be researched.
If equation has autocorrelation, we must remove it (Cochrane-Orcutt) or adjust for (use GLS instead of OLS).
I have now prepared VIDEO LECTURES on some of these mistakes, which explains the issues in MUCH MORE DETAIL. Furthermore, we are in process of writing up lecture notes for these lectures and preparing a textbook. These lectures, and associated materials, are availble from this link:
Applied Regression Analysis: Common Mistakes and How to Avoid Them
Choosing Topics for ECONOMETRICS theses - students are frequently confused about how to choose a topic for an econometrics thesis - whether M Phil or Ph.D. -- they often take topics which belong to economics, and not econometrics. An econometrics thesis will always have a different focus from an economics thesis. This lecture explains the essential differences.
AVOIDING Nonesense Regressions:
This involves tracing the causal chain: If X causes Y which causes Z then a regression of Z on X is not correct. One has to draw a PATH DIAGRAM with the causal chain and use this to run the regression. If some of the variables on the path are not directly observable, then one must use Structural Equation Modelling methods.
Many apparently sensible equations do not make sense because they ignore the idea of causation and mechanism. For example, suppose I want to investigatge the effect of culture on grwoth. I CANNOT do this by taking an equation for GNP and inserting some variable related to cultue in the equation. I have to first think about HOW culture would affect growth. WHAT are the variable that would be affected by culture? For example, we might consider work effort -- how hard workeres nwork -- as the primary variable -- then we would reun regressions of culture on work effort. Later we would relate work effortn to GNPl. This is how we workn through the CAUSAL mechansim by which culture affectgs growth. We cannot omit consdieration of the mechansims and come up with a valid analysis. This FAILURE to consider the mechansim and to estimate using causal chains is one of tghe most common econometric mikstakes.
Further guidance on these issues is provided in: