Regression Engine

Answers Whether a statistical relationship in your data is defensible: ordinary least squares with the diagnostics an opposing expert would run, robust standard errors where they are warranted, and the result framed against the admissibility criteria.From Your data only. The tool holds no dataset of its own; the Mohan and Daubert thresholds it applies are labelling conventions, registered in the panel.Not It establishes association under the model's assumptions, not causation, and it does not replace the expert who will re-run and defend the analysis — it is the preparation for that conversation.Use Preliminary. The automated figure is a first assessment; for a demand, a mediation brief or a report, a NorthLaw econometrician reviews and signs it as a fixed-fee reviewed assessment, or takes the full engagement.

Model selection, assumption testing, and admissibility analysis under the Canadian Mohan / White Burgess framework. The statistical diagnostics (heteroscedasticity, autocorrelation, normality, multicollinearity) address reliability concerns common to Daubert jurisdictions, but the legal criteria and citations are Canadian. Upload data, select your model, and get court-ready regression output with diagnostic flags.

OLS / WLS / Robust ARIMA 5 Diagnostic Tests Daubert Grades Auto Model Select

Every number below was chosen rather than measured: it is a modelling judgment, a labelling threshold, or a statistical convention. Each one was tested across the range shown in the September 2026 audit, and the effect on this tool's outputs is recorded in the methodology. Measured inputs — FJC federal civil case outcomes, SSA life tables, BLS benefit rates, BEA regional price parities, work-life expectancy tables — are not listed here; they carry their own provenance.

assumption	value	basis	range tested
`reg.mohan_r2_strong` R² above which (with a significant F) the fit is called adequate for Mohan/Daubert	0.3	judgment label	[0.2, 0.5]
`reg.mohan_r2_weak` R² above which the fit is called weak rather than none	0.1	judgment label	[0.05, 0.2]
`reg.mohan_n_adequate` observations at or above which the sample is called adequate	30	convention	[20, 50]
`reg.mohan_n_minimal` observations at or above which the sample is called minimal	15	convention	[10, 20]
`reg.dw_band` Durbin-Watson band treated as no autocorrelation	[1.5, 2.5]	convention	[[1.7, 2.3], [1.3, 2.7]]
`reg.vif_max` VIF at or above which multicollinearity is flagged	10	convention	[5, 10]
`reg.cooks_influential_frac` share of observations allowed above the Cook's distance threshold before flagging	0.05	judgment	[0.02, 0.1]
`reg.timeseries_min_days` date span above which data is treated as a time series	180	judgment	[90, 365]
`reg.timeseries_min_rows` rows required to treat dated data as a time series	12	judgment	[8, 24]
`reg.monthly_max_rows` rows at or below which a time series is called monthly rather than daily	120	judgment	[60, 240]
`reg.small_sample_n` rows below which the sample is flagged small	30	convention	[20, 50]
`reg.normality_small_n` residual count below which a normality failure is a higher Daubert risk	100	judgment	[50, 200]

Source of truth: assumptions.py in the NorthLaw repository (Audit C, F-56–F-59). If one of these does not fit your matter, the expert-review engagement re-runs the analysis with your figures.