Module 10 Activity

Scenario

Model a binary outcome and report it in probabilities a reader can use.

What you build

A logistic regression with odds ratios, predicted probabilities and a justified threshold.

Steps

1. Confirm the outcome is binary and check what a linear model would predict at the extremes of your predictor.
2. Fit with smf.logit and verify the output reports log-likelihood and a z statistic.
3. Exponentiate the coefficients and the bounds of conf_int(), label them as odds ratios, and give the baseline rate.
4. Produce predicted probabilities with get_prediction(grid).summary_frame() so the intervals stay in [0, 1].
5. Report the confusion matrix against the majority-class baseline and say how many positives were caught.
6. Check calibration by binning predictions against observed rates, then choose a threshold from the relative cost of misses and false alarms.

Evidence to hand in

• What a linear model predicted, and where it went out of range.
• The fitted model with its log-likelihood.
• The odds-ratio table with the baseline rate.
• The predicted-probability table with intervals.
• The confusion matrix against the majority baseline.
• The calibration check and your threshold justification.

Review checklist

• smf.logit was used, not smf.ols.
• Odds ratios are labelled as such and the baseline rate is given.
• Predicted-probability intervals came from summary_frame(), not from converting endpoints by hand.
• Accuracy is compared to the majority-class rate, not to 50%.
• The threshold is justified by costs, not left at 0.5 by default.

Mark complete