How to interprate the values of "t" received in the logistic regression model result ? Larger or smaller t-values with positive and negative sign..how they differ and what massage they convey..?
for logistic regression, we generaly focused on p-value ( > or < 0.05), rather than (z-statistics or t-statistics) for check signifcation parameters. so, if the corresponding p.value of a parameter is < 0.05 we say that is significatly different of zero. (we can fixed alpha at : 0.1 , 0.01 ).
for the estimated parameter we just interpret the positive or negatif effects on output variable based just on their signs. and we also use the odds ratio to make a sense of interpretation (as in linear regression)
You explained well, howevewr, I mean to ask how actually t-statistics is meaningful here and what massage it conveys...Larger or smaller t-values with positive and negative sign..how they differ and what massage they convey..?
the z-score is estimated as : Coefficient / Standard error, we know that the SE is always positive so the z-score take the sign of Coefficient, and the value of z-score is influenced by the Coefficint value and SE, if Coefficient is so big and SE is saml so we have a big z-score ...
Larger samples will give smaller standard errors and therefore higher Z-values, and smaller p-values, given equal coefficient values. Thus, Z-values and p-values do just say something about generalizability from sample to population (which of course also is dependent of the magnitude of the coefficient). Use OR-values to say something about how “influential it is or its contribution”.
It should be the Wald's test statistic. Wald's test tests if the effect is different from a null effect or not; in the case of a logistic regression, it means that it looks if the estimated regression coefficient B is different from 0 (or, equivalently, if the exponentiated coefficient -OR- is different from 1). As any t statistic, it is computed by dividing the distance between the estimated effect and the null effect by the SE.
It is important to divide between statistically significant (Wald, SE, p-value, Bayes CI) which means generalizability from sample to population, and substantial significance (e.g. effect size, R-square, OR). We should of course be interested in both aspects.
to interpret the t-value in the logistic regression, is by taking a look at the related parameter alone, and not necessarily by taking a look at its odd ratio