R + ztable: what are lcl and ucl? (logistic regression)

1
votes

ztable produces nice zebra (alternating striped) tables, useful for presenting dataframes and model objects (lm, glm, etc). However it's not clear to me what lcl and ucl are. The documentation says

‘ztable()’ shows the odds ratio(OR) and 95% confidence interval

But Googling indicates lcl/ucl to be lower/upper control limits, which seem to be like confidence intervals constructed with +/- 3 standard deviations from the mean.

Most confusingly, the lcl/ucl constructed by ztable don't always contain the coefficient estimate:

ztable(glm(factor(am) ~ disp, family = binomial(link = "logit"), data = mtcars))

So what do the lcl/ucl mean?

1
rlogistic-regression
The coefficient estimate is in parameter space. I'm imagining that the lcl and ucl's are in the measurement space. So I cannot see no reason that a control interval should include any coefficient. Seems for appropriate for stats.stackexchange.com – IRTFM
seems like lcl/ucl are the abbreviation the author uses for upper/lower confidence limit. So they are confidence intervals, so you would expect the coefficient to fall between the limits here is the funcrion – user2957945
@42- what do you mean by measurement vs parameter space? Both confint() and confint.default() applied on the model give intervals that contain the coefficient estimate, as I would expect. – adatum
@user2957945 thanks for the code. It seems the lcl/ucl are the confidence interval on the OR (odds ratio) and not on the coefficient estimate (log odds) ... – adatum
The parameters for logistic regression are the difference in estimated logodds between the baseline (intercept) and the non-base values of 'disp'. To figure out either probabilities or odds you need to put in values for 'disp' and include both the intercept baseline and add the difference. The confidence intervals from confint apply to the estimate of the difference on the log-odds scale, not on any probability or even to an odds estimate. – IRTFM

1 Answers

1
votes

From examining the source code for ztable.glm (thanks to @user2957945), the answer seems to be simply:

‘ztable()’ shows the odds ratio(OR) and 95% confidence interval (lcl, ucl) of the odds ratio

The odds ratio is the exponentiated coefficient (which is the logit or log odds), and the lcl/ucl are obtained by exponentiating the confidence interval of the coefficient.