OLS Regression problems assumptions

[MIGUEL IPANAQUE]

Hello dear, I am doing a study of factors associated with a my dependent variable (score obtained from a quality of life instrument), I use the independent variables: Age, Sex, educational level, Marital status, score on instrument of depression and score in psychological welfare instrument.

My problem is when I apply the univariate linear regression of: score of quality of life with civil status, I report p value of the test of significant significance but a category of this independent sample significance. Should I enter the multivariate linear model? Thank you very much for your comments.

[Dave Airey]

The regression is using an indicator variable for each level of civil status and comparing each level to a reference level which here is level 1. The p-values are for each level compared to level 1. If you wanted at p-value for the civil status as a whole, in this model show it is the F statistic p-value shown in the upper right. You could have also found the same p-value using a post estimation command like testparm or contrast, and you could have also used ANOVA instead of regress.

[MIGUEL IPANAQUE]

Thanks for your time Dave, I understand that the F test reports the global significance of whether or not it improves the variable to the model (p value (F) = 0.1836> 0.05), but my doubt persists, because it is reviewed as a previous step to these assumptions and does not report significance, but since the relation of the value reference (EC_2 = 0.043 <0.05), is it still advisable not to comply as a variable set? (F) is this maintained for multiple linear regression?

I know that it is also a way to do the bivariate by ANOVA and then the multiple linear regression, but in this case I first apply an unadjusted model (simple linear) and then an adjusted one (linear multivariate)

[Carlo Lazzaro]

Migule:
I find your last post difficult to follow.
As an aside to Dave's helpful insight, the F-test of your regression lacks of statistical significance: it means that your model does not have more informative power that the mean of of your dependent variable.
I think that a wiser methodological choice would consider including all the predictors consistent with the data generating process,
As an aside, why bothering with -xi:-instead of taking advantage of -fvvarlist-?

[Dave Airey]

I see. I did not quite understand your question. You are trying to decide on whether to include covariates in a multiple regression model and you are screening by univariate regressions and you notice a "conflict" between the t statistics and F statistic. If you corrected for multiple tests you would not have significance for the t-tests. On the other hand, if you think there is a scientific rationale for a possible relation between your potential variables and dependent variable, I would include them regardless.You seem to have hundreds of samples and you mention only a handful of possible covariates. You probably can afford them.