A yes-or-no outcome and a set of predictors. What an odds ratio actually says, the failure mode that produces impossible coefficients, and how to report it in APA 7.
Logistic regression predicts a binary outcome: passed or failed, retained or left, event or no event. It cannot use ordinary regression because a straight line will happily predict a probability of 1.4, which does not exist.
Instead it models the log odds of the outcome as a linear function of the predictors. Odds are the probability of the event divided by the probability of no event, and taking the logarithm makes the relationship linear and unbounded, which is the trick that makes the model work.
The cost is interpretability. A raw coefficient is a change in log odds, which means nothing to most readers. Exponentiating it gives an odds ratio, which means something but not what people usually assume.
An odds ratio is not a risk ratio. An odds ratio of 2.0 does not mean the outcome is twice as likely. It means the odds double, which is a different quantity.
The gap between them depends on how common the outcome is. When the event is rare, under roughly 10%, the two are close and the loose reading is nearly harmless. When the event is common, they diverge badly. An odds ratio of 2.0 on an outcome occurring 50% of the time corresponds to a risk moving from 50% to 67%, not to 100%.
This is one of the most frequent misreadings in applied research, and it usually inflates the reported effect. If your audience needs a probability, report predicted probabilities at meaningful predictor values alongside the odds ratios.
An odds ratio of 1.0 means no effect. Confidence intervals for odds ratios are therefore judged against 1, not 0, and an interval spanning 1 indicates a non-significant predictor.
Separation occurs when a predictor perfectly, or almost perfectly, predicts the outcome. Every case above a certain value experienced the event, and every case below it did not.
The model cannot converge. The likelihood keeps improving as the coefficient grows, so the estimate runs toward infinity. What you see is an enormous coefficient with an enormous standard error, an odds ratio in the thousands, and a p value that is meaningless.
It is easy to mistake for a spectacular finding. It is usually a sparse cell: a category with few cases where the outcome happened to be uniform. Quanta issues separation warnings rather than returning a huge coefficient without comment, which is the behavior that lets an implausible result reach a manuscript.
The remedies are to collapse the sparse category, remove the predictor, or use a penalized method designed for the situation. What is not acceptable is reporting the number as though it were an estimate.
Lay the data out one row per case, with the binary outcome in one column and predictors in others.
Upload the outcome and categorical predictors as text rather than numeric codes, so categories are typed as categories.
Check which outcome level is being modeled. Every coefficient's direction depends on it, and reading the model backwards inverts the entire interpretation.
Configure the outcome and predictors in the Setup panel on the right and the result appears in the center panel.
Count the rarer outcome, not the sample. A common guide is at least ten cases of the less frequent outcome per predictor. A study of 500 people with 20 events supports about two predictors, not twenty.
Report odds ratios with confidence intervals, and state which outcome level was modeled.
Higher perceived autonomy was associated with reduced odds of leaving within twelve months, OR = 0.62, 95% CI [0.44, 0.87], p = .006. The model was fitted to predict leaving.
Give the confidence interval for the odds ratio, not for the log odds, and remember the null value is 1.
Report the number of events, not only the sample size. A reader cannot judge whether the model was adequately powered without it.
Where the outcome is common, add predicted probabilities. They are what a non-specialist reader will take away, and they prevent the odds-ratio-as-risk-ratio misreading.
Every result carries a ReliCheck Intelligence card with one action: explain this result in plain language. It runs entirely on your Mac with Apple Intelligence. No upload, no API key, nothing to configure. The model only explains the result Quanta has already computed and validated. It never computes a statistic, and it never applies a decision on your behalf.
From the action bar you can add the result to a report, copy the table, or copy the APA text. Exported reports carry the APA table, its note, and a caution specific to that analysis beneath it.
Quanta's logistic, Poisson, and ordinal regressions are validated against R's glm and MASS::polr, with coefficients matched to 1e-3. The engine is pure Swift with no R or Python at runtime. Full record on the validation page.
Regression is part of every Quanta subscription, $9.99 a month or $79.99 a year after the 30-day trial. One price for everyone. See pricing.
The analysis, the report builder, and the on-device explanations all work with the network off. Your dataset is never uploaded.
Multiple linear regression for a numeric outcome · Ordinal regression for an ordered outcome with more than two levels · Chi-square for two categorical variables without controls · the complete list on the analyses page.