Methods

How the numbers are verified

Every model on this site is fitted in TypeScript, in your browser or at build time. Those ports are only useful if they agree with the R code I submitted, so each one is tested against reference output regenerated from the original assignment code.

The pipeline

1
Re-run the original R
scripts/reference.R repeats the model calls from the Assignment 2 and 3 R Markdown files verbatim and writes every reproduced number at full precision to a JSON fixture; scripts/diagnostics-reference.R does the same for the 2026 model checks.
2
Port the algorithms
IRLS mirrors glm.fit line by line (starting values, Householder QR, convergence rule); multinomial, proportional-odds and GEE fits solve the same likelihood or estimating equations with Newton steps.
3
Test for parity
A vitest suite compares the ports with the fixture and with numbers printed in the original notebooks. CI runs it on every push; the table below is the same comparison, computed when this page was built.

Parity table

49 of 49 checks within tolerance. Reference: R version 4.6.1 (2026-06-24); nnet 7.3.20, MASS 7.3.65, geepack 1.3.13; for the 2026 model checks also VGAM 1.1.14, brant 0.3.0 and lme4 2.0.6. Differences are relative (absolute for values below 1).

Beetle

β₀ (straight line)
R
-60.10327943
TS
-60.10327943
Δ
9.0e-15 ≤ tolerance 1.0e-9
β₁ (straight line)
R
33.93416481
TS
33.93416481
Δ
7.7e-15 ≤ tolerance 1.0e-9
SE(β₁)
R
2.902867031
TS
2.902867031
Δ
4.3e-15 ≤ tolerance 1.0e-9
Residual deviance
R
13.63338416
TS
13.63338416
Δ
3.0e-15 ≤ tolerance 1.0e-10
Pearson X²
R
12.11328428
TS
12.11328428
Δ
1.1e-14 ≤ tolerance 1.0e-9
AIC
R
43.83142238
TS
43.83142238
Δ
9.2e-15 ≤ tolerance 1.0e-10
LD50
R
1.771173087
TS
1.771173087
Δ
3.0e-15 ≤ tolerance 1.0e-10
P(killed | dose 1.8)
R
0.7267543314
TS
0.7267543314
Δ
4.9e-15 ≤ tolerance 1.0e-10
β₂ (quadratic)near-collinear design
R
153.5606193
TS
153.5606193
Δ
2.5e-13 ≤ tolerance 1.0e-7
Deviance (quadratic)
R
5.106974528
TS
5.106974528
Δ
2.5e-14 ≤ tolerance 1.0e-9

Quantity	R	TypeScript	Difference	Tolerance
β₀ (straight line)	-60.10327943	-60.10327943	9.0e-15	1.0e-9
β₁ (straight line)	33.93416481	33.93416481	7.7e-15	1.0e-9
SE(β₁)	2.902867031	2.902867031	4.3e-15	1.0e-9
Residual deviance	13.63338416	13.63338416	3.0e-15	1.0e-10
Pearson X²	12.11328428	12.11328428	1.1e-14	1.0e-9
AIC	43.83142238	43.83142238	9.2e-15	1.0e-10
LD50	1.771173087	1.771173087	3.0e-15	1.0e-10
P(killed \| dose 1.8)	0.7267543314	0.7267543314	4.9e-15	1.0e-10
β₂ (quadratic)near-collinear design	153.5606193	153.5606193	2.5e-13	1.0e-7
Deviance (quadratic)	5.106974528	5.106974528	2.5e-14	1.0e-9

Attitudes

model.5 Education slope
R
-0.2635160023
TS
-0.2635160023
Δ
3.9e-16 ≤ tolerance 1.0e-8
model.5 Female × Education
R
-0.08172065599
TS
-0.08172065599
Δ
8.3e-17 ≤ tolerance 1.0e-8
model.5 deviance
R
20.24417710
TS
20.24417710
Δ
2.0e-14 ≤ tolerance 1.0e-9
model.4 deviance
R
15.15999289
TS
15.15999289
Δ
2.3e-15 ≤ tolerance 1.0e-8
model.6 deviance
R
20.23682242
TS
20.23682242
Δ
5.4e-15 ≤ tolerance 1.0e-8
model.7 deviancealiased columns re-coded
R
9.947281513
TS
9.947281513
Δ
5.7e-15 ≤ tolerance 1.0e-8

Quantity	R	TypeScript	Difference	Tolerance
model.5 Education slope	-0.2635160023	-0.2635160023	3.9e-16	1.0e-8
model.5 Female × Education	-0.08172065599	-0.08172065599	8.3e-17	1.0e-8
model.5 deviance	20.24417710	20.24417710	2.0e-14	1.0e-9
model.4 deviance	15.15999289	15.15999289	2.3e-15	1.0e-8
model.6 deviance	20.23682242	20.23682242	5.4e-15	1.0e-8
model.7 deviancealiased columns re-coded	9.947281513	9.947281513	5.7e-15	1.0e-8

Death penalty

[D][V][P] G²
R
137.9293556
TS
137.9293556
Δ
6.2e-16 ≤ tolerance 1.0e-8
[DV][P] G²
R
8.131610900
TS
8.131610900
Δ
2.0e-15 ≤ tolerance 1.0e-8
[DV][DP] G²
R
7.910160591
TS
7.910160591
Δ
3.4e-16 ≤ tolerance 1.0e-8
[VD][VP] G²
R
1.881895464
TS
1.881895464
Δ
1.2e-15 ≤ tolerance 1.0e-8
Logistic ~ Victim_Race deviance
R
1.881895464
TS
1.881895464
Δ
2.8e-15 ≤ tolerance 1.0e-8

Quantity	R	TypeScript	Difference	Tolerance
[D][V][P] G²	137.9293556	137.9293556	6.2e-16	1.0e-8
[DV][P] G²	8.131610900	8.131610900	2.0e-15	1.0e-8
[DV][DP] G²	7.910160591	7.910160591	3.4e-16	1.0e-8
[VD][VP] G²	1.881895464	1.881895464	1.2e-15	1.0e-8
Logistic ~ Victim_Race deviance	1.881895464	1.881895464	2.8e-15	1.0e-8

Pneumoconiosis

Severe vs normal slopevs multinom(reltol = 1e-15)
R
0.1092853327
TS
0.1092853322
Δ
4.1e-10 ≤ tolerance 1.0e-6
Multinomial deviance
R
417.4495633
TS
417.4495633
Δ
7.0e-13 ≤ tolerance 1.0e-8
SE(severe slope)nnet Hessian
R
0.01646978363
TS
0.01646976569
Δ
1.8e-8 ≤ tolerance 1.0e-5
Proportional-odds slopevs polr(reltol = 1e-15)
R
0.09590408742
TS
0.09590408815
Δ
7.3e-10 ≤ tolerance 1.0e-6
Cut-point θ₂
R
4.869048624
TS
4.869048624
Δ
3.3e-11 ≤ tolerance 1.0e-6
Proportional-odds deviance
R
416.9188336
TS
416.9188336
Δ
6.7e-14 ≤ tolerance 1.0e-8
SE(slope)polr uses a finite-difference Hessian
R
0.01193789093
TS
0.01193657086
Δ
1.3e-6 ≤ tolerance 2.0e-4

Quantity	R	TypeScript	Difference	Tolerance
Severe vs normal slopevs multinom(reltol = 1e-15)	0.1092853327	0.1092853322	4.1e-10	1.0e-6
Multinomial deviance	417.4495633	417.4495633	7.0e-13	1.0e-8
SE(severe slope)nnet Hessian	0.01646978363	0.01646976569	1.8e-8	1.0e-5
Proportional-odds slopevs polr(reltol = 1e-15)	0.09590408742	0.09590408815	7.3e-10	1.0e-6
Cut-point θ₂	4.869048624	4.869048624	3.3e-11	1.0e-6
Proportional-odds deviance	416.9188336	416.9188336	6.7e-14	1.0e-8
SE(slope)polr uses a finite-difference Hessian	0.01193789093	0.01193657086	1.3e-6	2.0e-4

Wheeze GEE

β age (exchangeable)vs geeglm(epsilon = 1e-14)
R
-0.1133849967
TS
-0.1133849967
Δ
3.7e-16 ≤ tolerance 1.0e-9
β smoke (exchangeable)
R
0.2650757406
TS
0.2650757406
Δ
2.8e-15 ≤ tolerance 1.0e-9
Robust SE smoke
R
0.1777465511
TS
0.1777465511
Δ
1.3e-15 ≤ tolerance 1.0e-8
Working correlation α
R
0.3543049158
TS
0.3543049158
Δ
2.8e-16 ≤ tolerance 1.0e-9
Scale φ
R
0.9984646059
TS
0.9984646059
Δ
1.1e-16 ≤ tolerance 1.0e-9
Robust SE smoke (default fit)geeglm default epsilon = 1e-4
R
0.1777465499
TS
0.1777465511
Δ
1.3e-9 ≤ tolerance 1.0e-7
Naive GLM SE smoke
R
0.1234730601
TS
0.1234730601
Δ
2.8e-17 ≤ tolerance 1.0e-9

Quantity	R	TypeScript	Difference	Tolerance
β age (exchangeable)vs geeglm(epsilon = 1e-14)	-0.1133849967	-0.1133849967	3.7e-16	1.0e-9
β smoke (exchangeable)	0.2650757406	0.2650757406	2.8e-15	1.0e-9
Robust SE smoke	0.1777465511	0.1777465511	1.3e-15	1.0e-8
Working correlation α	0.3543049158	0.3543049158	2.8e-16	1.0e-9
Scale φ	0.9984646059	0.9984646059	1.1e-16	1.0e-9
Robust SE smoke (default fit)geeglm default epsilon = 1e-4	0.1777465499	0.1777465511	1.3e-9	1.0e-7
Naive GLM SE smoke	0.1234730601	0.1234730601	2.8e-17	1.0e-9

Model checks (2026)

Beetle leverage, dose 1.72hatvalues()
R
0.3491434715
TS
0.3491434715
Δ
2.6e-14 ≤ tolerance 1.0e-9
Beetle Cook's D, dose 1.76cooks.distance()
R
1.033666488
TS
1.033666488
Δ
6.3e-13 ≤ tolerance 1.0e-8
Quasi-binomial dispersion
R
2.018893941
TS
2.018893941
Δ
3.1e-14 ≤ tolerance 1.0e-9
Quasi F test p, quadratic term
R
0.03285800828
TS
0.03285800828
Δ
3.3e-15 ≤ tolerance 1.0e-9
Profile CI for slope, loweruniroot on offset refits
R
28.54450924
TS
28.54450924
Δ
1.2e-16 ≤ tolerance 1.0e-8
Profile CI for LD50, upper
R
1.778679926
TS
1.778679926
Δ
1.5e-14 ≤ tolerance 1.0e-9
LD50, quadratic logit: profile CI upperuniroot on constrained refits
R
1.788363698
TS
1.788363698
Δ
2.4e-15 ≤ tolerance 1.0e-9
LD50, cloglog lineMASS::dose.p()
R
1.778204508
TS
1.778204508
Δ
7.1e-11 ≤ tolerance 1.0e-8
Brant X², parallel slopesbrant package
R
0.05135868732
TS
0.05135781870
Δ
8.7e-7 ≤ tolerance 1.0e-5
Separate-slopes β₂VGAM::vglm
R
0.09607722320
TS
0.09607722390
Δ
7.1e-10 ≤ tolerance 1.0e-7
AR(1) working correlation αgeeglm(corstr = 'ar1')
R
0.4910089758
TS
0.4910089758
Δ
2.2e-15 ≤ tolerance 1.0e-9
AR(1) robust SE smoke
R
0.1811950126
TS
0.1811950126
Δ
1.1e-15 ≤ tolerance 1.0e-8
QIC, exchangeablegeepack::QIC()
R
1829.482937
TS
1829.482937
Δ
9.6e-15 ≤ tolerance 1.0e-9
GLMM β smokeglmer(nAGQ = 25) stops early
R
0.3985632954
TS
0.3985626077
Δ
6.9e-7 ≤ tolerance 2.0e-3

Quantity	R	TypeScript	Difference	Tolerance
Beetle leverage, dose 1.72hatvalues()	0.3491434715	0.3491434715	2.6e-14	1.0e-9
Beetle Cook's D, dose 1.76cooks.distance()	1.033666488	1.033666488	6.3e-13	1.0e-8
Quasi-binomial dispersion	2.018893941	2.018893941	3.1e-14	1.0e-9
Quasi F test p, quadratic term	0.03285800828	0.03285800828	3.3e-15	1.0e-9
Profile CI for slope, loweruniroot on offset refits	28.54450924	28.54450924	1.2e-16	1.0e-8
Profile CI for LD50, upper	1.778679926	1.778679926	1.5e-14	1.0e-9
LD50, quadratic logit: profile CI upperuniroot on constrained refits	1.788363698	1.788363698	2.4e-15	1.0e-9
LD50, cloglog lineMASS::dose.p()	1.778204508	1.778204508	7.1e-11	1.0e-8
Brant X², parallel slopesbrant package	0.05135868732	0.05135781870	8.7e-7	1.0e-5
Separate-slopes β₂VGAM::vglm	0.09607722320	0.09607722390	7.1e-10	1.0e-7
AR(1) working correlation αgeeglm(corstr = 'ar1')	0.4910089758	0.4910089758	2.2e-15	1.0e-9
AR(1) robust SE smoke	0.1811950126	0.1811950126	1.1e-15	1.0e-8
QIC, exchangeablegeepack::QIC()	1829.482937	1829.482937	9.6e-15	1.0e-9
GLMM β smokeglmer(nAGQ = 25) stops early	0.3985632954	0.3985626077	6.9e-7	2.0e-3