I could experience that eviews shows an error massage near singular matrix when I tried to perform a fixed effect model for gravity model with panel data. Then I checked my panel variables for multicoliniarity problem and omitted such variables.then I did fixed effect model again and still I exlerience the same error.but hausman test do reject null hypothesis and hence fixed effect model should be the best instead of random effect model.I am confused with that error massage and what hausman test says.can anybody help me to get rid of thisfish problem please
I have 107 counntries and 17years with 7 independent variables.I am going to check the determinants of sri lankan sea food export performance using gravity approach.I have some variables like distance between sri lanka and particular country which are do not vary within cross section.is this the problem??but the hausman test results is on favour of fixed effect model.I get an error message near singular matrix.if hausman test conclude the best will be FE model what should I do to overcome this problem.please help me
This has nothing to do with the N/T ratio - it will always happen with what you are trying to do.
The fixed effects for country will remove all variation that is to be explained so putting in a country variable cannot work as there is nothing to explain- fixed effects throw out the baby with the bathwater.
The only way to proceed is to use a random effects model in its within- between form or in a Mundlak specification. This specification does not eat up all the degree of freedom and you can include level 2 time-invariant country variables which you are interested in . Also the Hausman test is not telling that you should use fixed over random but that the cross-sectional (time invariant) between effect is different from the within time-varying effect; it is not a fault but a feature; you should be interested in that.
These are rather heterodox views see
https://en.wikipedia.org/wiki/Kelvyn_Jones
section beginning:
"He (with colleagues) has challenged the 'gold standard' that fixed effects should be the standard approach to the analysis of Panel data and that a Hausman test is an appropriate way of choosing between a Fixed effects model and a Random effects model. Somewhat controversially they argue that a particular form of the random effects model (the within-between model or the similar Mundlak model) offers all that fixed effects can provide and more"
and the papers cited there
Article Explaining Fixed Effects: Random Effects Modeling of Time-Se...
Article Fixed and Random effects models: making an informed choice
Article Understanding and misunderstanding Group mean centering: a c...
Thank you so much. I will go through this materials. Since I faced many problems with eviews I tried with STATA.STATA works for fixed effect model but it omits a very important variable. There also hausman test is favor for fixed effect model. But It omits very important variable. Since that situation can I select random effect model without considering hausman test?
Yes but you need to consider the within-between specification if you are interested in the time invariant variables; it is not really a software issue.
what do you mean by within-between specification sir? I do not have much knowledge
It is all set out in our in the earlier papers I cited; apologies for the shameless publicity! Apologies for not taking the time to reply fully but I have set four hares running today with what maybe seen as provocative replies, but that is what this site is good for. Have a look at the stuff and then come back, and I will have a go at explaining more.
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Graphs