By "P value", what do you mean? Do you mean the p-value associated with an individual predictor variable? The assertion in the second part of your question is controversial. Whether you include a particular variable in your model can be based on a number of reasons, but the p-value is likely one of the least important.

Sujal, you may find the attached paper of some interest. Best wishes, David Booth

@Daniel Wright, Yes P value associated with individual predictor variable. If its coming to P>0.05 is its really matters? In logistics regression analysis are we concern more with Sig value like t test or ANOVA or Odds ratio is enough? Thank you for showing interest.

Hi Sujal,

While the P-Value can indicate a potential relationship with a reasonable degree of reliability, it is not the main factor I would look at when selecting variables for inclusion. This measure can change dramatically depending on how the model is built across iterations, in my experience. Measures such as the Akaike Information Criterion (AIC) can be a good way of identifying variables for inclusion, as can measures such as The Variance Inflation Factor (VIF) to identify collinearity issues. I usually run analysis in R software, and there are easily applicable functions to assess in this manner. I often work with population data and finding instances of collinearity is often what primarily determines my variables for inclusion. Removing one variable that has associations with other independent variables can have a dramatic impact on the p-values associated with them, so I would always value this consideration more.

I hope you found some of above helpful, feel free to get in contact if you would like some information on anything mentioned. I am also certain that there are many people better versed on this topic than I am that may have alternative tools / suggestions!

Thanks & Regards,

William

The null hypothesis in linear regression is - this variable does not effect that variable. If p-value of less than 0.05, then you can reject the null hypothesis. It also means there is a statistically significant relationship between the two variables.

It is just to select the predictors. In logistic likelihood matters.

In logistic regression analysis, before performing a multivariate logistic analysis, we go through a series of madels but we rely on the p-value. Thus, once included variable by variable in the univariate logistic analysis we decide whether or not to keep the variables for the multivariate logistic analysis because any variable included in the univariate logistic analysis with a p-value greater than 0.20 must be excluded from the multivariate logistic model.