anyone remember how to prove simple heteroskedasticity between two variables? Does it matter if I have cross sectional or panel data?

I don't think you'll find many people knowledgeable in the subject. Perhaps you can find it in an SPSS manual or something?

I have it in my book but most of it is for proofing heteroskedasticity in relations to gauss-markov assumptions (for population linear regression or OLS-BLUE.) Was wondering how to do it on a simple two variable dataset

Well, I'm afraid I can't help you with any specifics as I was quite aware it would've taken me longer had I done a subject that needed all that for my dissertation

The only things I can think of is perhaps you can derive the answer when you try and find the relation between gauss-markov assumptions and the other type you're looking for. If you can fully appreciate how the proof applies for gauss-markov assumptions, then you can successfully translate it to other situations if applicable. So the question really, is: do you know just what heteroscedasticity is, how it is found and what its implications are?
Secondly, perhaps there's something found in the sources of the proofs in relations to gauss-markov assumptions

Sorry that I haven't got anything more than pre-Uni stuff for you, I've had no education in this subject When you found an answer let me know, I'm interested in learning about it

Could always plot it and estimate best fit line and see if the error rate is relatively constant or not.



This is simple heteroskedasticity since the error rate from the best fit line is not constant throughout the graph. Homoskedastisticity would be if it held the same error range throughout the graph.

This post was edited by timmayX on Jul 3 2017 01:22pm

well I was thinking of a way to prove it in terms of value.

my issue is with non-linear lines such as


so far I have tried to run anova on it (there were a lot more variables, but I included "weight loss" to show my partner that colinearity occurs when you have a variable that is a function of given variables, and it looks really amateurish to include this)


So to make the graph become homoskedastic, from experience, I know you apply a dummy variable (in this case ln) to increase the R-squared value and reduce the variation in the regression. However, I forgot how to prove the graph is heteroskedastic (in math/stats terms) to apply this. For OLS "Best Linear Unbiased Estimator" we use one of gauss-markov assumptions before you do a regression, and prove it is homoskedastic by covariance over variance equation (or it could be the other way around.)

So I am curious if anova can be run on the heteroskedastic data and proofed through least sums estimator? Or proofed that it cannot be run with it.

Was wondering how it would apply in a given equation like :

part of the gauss-markov assumptions of linear regressions.