Regression

[qim]

I wonder if you can help me with the following. I am not a statistics student but need to interpret some tables. I asked for some help a few days ago but as I did not get a reply I got some books and I am trying to understand what is going on...

The table below is supposed to be a regression model showing what happens in a choice between males and females (gender male=1; female = 0) and (ice = 1; no ice = 0)

From what I can read when the choice is ice there is a negative coefficient implying that females prefer ice (cream) just slightly more than males.

Then, and this is the difficult bit, the p-value being -1.39 says that this is not a significant statistic. But that goes against earlier tables that agree with these results.

Can yo help me? Am I totally off the mark?

Thank you

qim


. logit con gender

Iteration 0: log likelihood = -1595.2709
Iteration 1: log likelihood = -1594.3081
Iteration 2: log likelihood = -1594.3079

Logistic regression Number of obs = 2756
LR chi2(1) = 1.93
Prob > chi2 = 0.1652
Log likelihood = -1594.3079 Pseudo R2 = 0.0006


ice Coef. Std. Err. z P>z [95% Conf. Interval]

gender -.1204865 .0869281 -1.39 0.166 -.2908625 .0498895
_cons -.9632477 .0576305 -16.71 0.000 -1.076201 -.850294

[jahred]

Then, and this is the difficult bit, the p-value being -1.39 says that this is not a significant statistic. But that goes against earlier tables that agree with these results.

Can yo help me? Am I totally off the mark?

the p-value is a probability, and as such must lie between 0 and 1. so it can't be negative, and even if it were, its absolute value couldn't be greater than 1.

the p-value associated with the gender variable looks to me to be .166, so it still looks as if the gender variable is not particularly significant in this model.

[qim]

Thank you jahred

Of course, the p-value is 0.166...

How ould I read the chi2 values, please?

Thanks

[jahred]

iirc the LR chi^2(1) value refers to the test statistic -2 log L, where L is the appropriate likelihood ratio (LR) for a likelihood ratio test.

for a 2x2 contingency table (i.e. male/female ice/no ice), -2 log L has a chi-square distribution with one degree of freedom, hence the chi^2(1).

i believe here it's used for the test

Ho: gender & ice cream preference are not associated
Ha: gender & ice cream preference are associated

so it is equivalent to testing the coefficient for gender in your logistic regression.

notice that the p-values are the same; P(-2 log L > 1.93) = 0.166. so for the likelihood ratio test above, there is insufficient evidence to reject Ho.