stats_discrete: 7 tools.
| Tool | What it does | Notes |
|---|---|---|
stats_logit | Binary choice model for a 0/1 outcome (default or not, up or down, churn): logit or probit by maximum likelihood. | core |
stats_multinomial_logit | Multinomial logit for an unordered categorical outcome with three or more values (choice of venue, rating bucket, sector): one coefficient set per outcome against the lowest (base) outcome. | |
stats_conditional_logit | Conditional maximum likelihood within groups: McFadden’s conditional logit for choices among alternatives (which venue, broker or mode is chosen given each alternative’s attributes) and fixed-effects logit, Poisson or multinomial logit for panels, where each group’s own intercept is conditioned away (no intercept is estimated). | up to 180 s |
stats_ordered_model | Ordered logit or probit for a ranked outcome (credit rating notches, analyst recommendation 1-5, survey scale): one slope per regressor and a cutpoint between each pair of adjacent categories, no intercept. | up to 120 s |
stats_count_regression | Regression for counts (trades per day, defaults per portfolio, claims per policy): Poisson, negative binomial (NB2, NB1, geometric or NB-P) or generalized Poisson, with optional offset or exposure for rates. | |
stats_zero_inflated | Zero-inflated count model: a mixture of structural zeros (rows that can only be zero, such as dormant accounts) and a Poisson, negative binomial or generalized Poisson count. | up to 120 s |
stats_hurdle_truncated | Hurdle model (one process decides zero or positive, a zero-truncated count gives the size: whether a client trades, then how many trades) or left-truncated count model (counts at or below the truncation point are never recorded). | up to 120 s |
fincept_describe_tool.