INTERPRETING This indicates the number of

lOGISTIC REGRESSION ANALYSES individuals included in the

analysis, which may be smaller
When fitting a logistic regression model, the first than the dataset's total rows due
variable specified is the outcome. Continuous variables to row-wise deletion. Only
Odds ratios are always greater than 0, with 1
(e.g., age) are measured per unit increase, with each individuals with complete
representing the null value. Values above 1 indicate
value serving as a reference for the next higher value. information on the outcome and
a positive association, while values below 1 reflect an
For categorical variables, you can select the reference all predictors are included,
inverse association. As implied by the term “ratio,” a
group, which can be the lowest (i.var), highest meaning those with any missing
comparison is always involved; omitting this
(ib(last).var), or a specific category (ib(n) in Stata). values are excluded.
comparison leads to incorrect interpretations.
This assesses whether this
Let's break this down:
model is better than nothing
Log-likelihood is the log
(i.e., a model with no
of the likelihood Just another way of capturing
predictors). The null hypothesis
(probability) that the uncertainty in measurements.
states there is no difference
model perfectly predicts We can use this to get the
between this model and an
the data. The highest confidence intervals (CI):
empty one. The alternative
value any probability Lower CI: odds ratio - (1.96 * SE)
hypothesis is that this model
can take is 1, and since Upper CI: odds ratio + (1.96 * SE)
differs from an empty model.
log(1) is 0, a log-
With the small p-value < 0.001
likelihood of 0 implies a
here, we can safely reject the
perfect fit. Values closer
null hypothesis and conclude
to 0 suggest a better fit.
that this model with predictors
This metric is best used
is significant different!
to compare models.

The R here stands for

"Residual". So the R2 tells us
The Z statistic is like a how much of the total
standardized tape rule that variance in the data is
measures distances in explained by the model,
standard deviations We indicating the proportion of
need a standardized scale variability that remains
because different variables unexplained. The true R2 is in
can be in various units (e.g., linear regression, we use the
kilometers, inches, or log
odds in logistic regression).
. pseudo R2 here because the
underlying assumptions are
The Z statistic tells us how different. The number here is
many standard deviations a a proportion so we need to
value is from the mean (the multiply by 100 to turn it into
mean is calibrated to 0 on a percentage
the Z scale). The farther a
value is from the mean (or
zero), the stronger the
The constant (intercept) indicates the odds of the outcome when all The p-value and confidence intervals indicate
effect in the analysis.
predictors are at their reference levels or 0, compared to those with at statistical significance, but prefer the 95% confidence
least one non-baseline value. In this case, it reflects the odds of interval as it also reflects estimate precision. Wide
reporting weight gain among participants who did not quit smoking, intervals suggest small sample sizes, while p-values
were aged 0, had no education, and were sedentary. However, this less than 0.05 indicate significance. Confidence
constant has limited real-world relevance, as a value of 0 is nonsensical intervals that exclude 1 are also significant. Both
for many indicators, such as height or weight. measures convey the same information.