Truncated regression
Regression · Method and input reference
Rows are independent observations. No missing values. Truncated and Tobit models use a lower bound and Gaussian errors. Exponential fits log(y) by OLS. Logistic requires a binary 0/1 outcome. Synthetic example: Assay responses observed only above zero, then standardized exposure; unobserved values below the threshold are excluded.
Open calculator · Compare methods in the decision matrix
Required inputs
- Data: outcome first, then predictor columns — Numeric observations; preserve the row and column structure described above.
- Lower truncation / censoring bound — Enter the method parameter described above.
Review missing values and data coding before running. Sample inputs are synthetic teaching examples.
Implementation
Lifescio custom lower-bound Gaussian likelihood, optimized with BFGS; numerical observed-information standard errors from statsmodels.tools.numdiff.approx_hess.
- scipy 1.15.3: scipy.optimize.minimize, scipy.stats.norm.logpdf, scipy.stats.norm.logcdf, scipy.stats.norm.logsf
- statsmodels 0.14.4: OLS, tools.numdiff.approx_hess
These are pinned library versions; saved results record the versions actually used at runtime.