Examples

Each page here follows one of the scripts in examples/ of the repository: it explains what the script shows, with its output, and ends with the script itself, which you can download from the link at the top of the page. The scripts run on simulated data from logit_stream.simulate, so there is nothing else to download and nothing to configure. Each script stands alone, so a downloaded one runs anywhere logit-stream is installed:

pip install logit-stream
python quickstart.py

or, from a clone of the repository, python examples/quickstart.py. Most take a few seconds; the random-intercepts and mixed-logit examples take about a minute, since they fit several models each. The scripts that write files put them in an output/ directory next to the script, which you can delete afterwards. The outputs shown on these pages are from those runs, abridged; your numbers may differ in the last digits.

Quickstart	every estimator once, and the reporting and prediction methods they share. Start here.
Binary outcomes	the logit and probit with fixed effects, and Firth’s penalized likelihood for separation
Multinomial logit	parameterization, standard errors (robust, clustered, multi-way, jackknife), weights, ridge
Ordered logit and probit	cutpoints, levels, residuals and predictions
Conditional logit	long-format choice data, alternative-specific constants, policy scenarios
Nested logit	nests of alternatives, the dissimilarity parameters, substitution, deeper trees
Mixed logit	random coefficients, panels of repeated choices, simulation draws
Fixed effects	absorbing a fixed effect in the multinomial and ordered logit, and several
Random intercepts	group intercepts with an estimated covariance, nested groupings, shrinkage
Incidental parameters	the fixed-effects logit’s bias in short panels, the bias correction, and Chamberlain’s conditional logit
Imputation at scale	predicting a characteristic for a large file, and combining the predictions with other information

The simulated data

logit_stream.simulate ships with the package, so you can try the estimators, or check your own code, on data whose answer you know (see the reference):

• simulate_people is a population shaped like the demographic imputation problem: people in PUMAs nested in states, with age, sex, income and household characteristics, and three outcomes drawn from logit models of them – Hispanic origin (binary), race and ethnicity (six categories) and educational attainment (five ordered levels) – each with PUMA effects. truth_people returns the coefficients it draws from.

• simulate_choices is long-format mode-choice data (train, metro, car), in the shape of the Swissmetro survey, drawn from the conditional logit; with nests= from the nested logit, and with tasks= and sd_time= from the mixed logit with a panel of repeated choices.

• simulate_panel is a panel with a binary outcome and unit effects correlated with a covariate, for the fixed-effects logit.