Testing homoscedasticity for multiple regression in SPSS

Question

I am conducting a multiple regression with 1 DV and 6 IVs. I am trying to test Homoscedasticity on SPSS using a scatterplot since all my variables are scales. I conducted a the residual vs predictor value scatterplot and I think it might be a little heteroscadestic.

How do I know which variable is the one causing the problem? And what should the next step be to try to make my data homoscedastic?

This question appears to be off-topic because it is about how to do something in SPSS. — gung - Reinstate Monica, Feb 22 '14 at 16:58
I would like to know how to tell if there is homoscedasticity. I managed to issue a scatterplot between residual and predicted value. But I have no idea how the scatterplot should look like. — dissertationhelp, Feb 22 '14 at 17:01
Welcome to the site, @dissertationhelp. If you are only asking how to do this in SPSS, this Q would be off-topic for CV (see our help center). You can find some SPSS resources at the tag's wiki. If you have a question about the substantive statistical issues here, please edit to clarify, if not, this Q may be closed. — gung - Reinstate Monica, Feb 22 '14 at 17:04
If you just want to know what the plots should look like, you can see examples in this thread: What does having constant variance in a linear regression model mean? — gung - Reinstate Monica, Feb 22 '14 at 17:06

score 2 · Answer 1 · answered Mar 09 '16 at 11:48

In regression analysis, residuals should be independent from response variable, all of the predictors as well as the predicted value of response variable. You can detect, if there is any pattern in these plots in SPSS using these steps:

Analyze > Regression > linear > plots [Zresidual vs Zpredicted and zresidual vs dependent]. It is also better to plot Zresidual Vs all predictors.

Now if the assumption of homoscedasticity is violated, then you can use regression with WLS weights. To compute weights in SPSS:

Analyze > Regression > weight estimation > select dependent & independent variables (SPSS use these names for response and predictors) > select weight variable for which hetroscedasticity is detected. The default power range is -2 to 2 by 0.5 in SPSS.> Click Ok > read the power for which log likelihood is maximize

Repeat the test few more time by narrowing the range with smaller increment to get better weight and save variable by using options in weight estimation.

Rerun regression using these weight.

Testing homoscedasticity for multiple regression in SPSS

1 Answers1