What do I use
Economics, finance and accounting
Firm performance, market studies, policy evaluation, organisational research.
Your question, and what answers it
The Hausman test decides, alongside the poolability F and the Breusch-Pagan LM. Report all three.
In ZStat: open your study, go to Analysis, and choose panel regression.
With multicollinearity, heteroscedasticity and Durbin-Watson reported, because a reviewer will ask.
In ZStat: open your study, go to Analysis, and choose ols regression.
With the ordering justified by design rather than by fit.
In ZStat: open your study, go to Analysis, and choose mediation.
With the classification table and the base rate beside the accuracy.
In ZStat: open your study, go to Analysis, and choose logistic regression.
Split chronologically. A model validated on shuffled data trains on next month to predict last month.
In ZStat: open your study, go to Analysis, and choose arima.
Or its non-parametric equivalent, because financial ratios are rarely normal.
In ZStat: open your study, go to Analysis, and choose one way anova.
What goes wrong in economics, finance and accounting
- Adopting fixed effects with no Hausman statistic. That asserts a modelling decision rather than making one.
- Durbin-Watson computed across the boundary between one firm and the next, which measures the sort order rather than autocorrelation.
- Judging a cross-sectional model by R-squared. A value of .10 can accompany a coefficient that matters.