A telecom carrier's churn model is a logistic regression, and the positive class is churn. The coefficient on the number of support calls is positive. What does that positive coefficient mean?
Select an answer to reveal the explanation.
Short Explanation
A logistic coefficient is a shove on the log-odds scale, not a coupon worth that many probability points. A positive shove raises the log-odds of churn, which is your positive class. The trap is reading the sign as protection or as a raw probability jump.
Full Explanation
In logistic regression the linear predictor is the log-odds of the positive class. A positive coefficient on support calls means that, with the other covariates held fixed, a higher call count increases that log-odds and therefore increases the modeled chance of churn. The increase is not a fixed number of probability points equal to the coefficient, because the logistic function curves and the same log-odds step maps to different probability steps depending on the current level. Reading the positive sign as a protective factor reverses the direction: a positive coefficient raises, rather than lowers, the log-odds of the positive class. Saying the count changes only the intercept is also wrong. The intercept is the baseline log-odds, and the support-call coefficient is the slope on that feature. Exam caveat: the direction is for the class coded as positive, so a model that codes staying as the positive class would flip the story. Operational check: increase the support-call value on a scored customer while holding the other inputs fixed, and confirm the predicted log-odds of churn move up.