# I replicated ordered probit output (using a Stata dataset) with four R packages, compared it to Stata, unable to even come close to the Stata output

I was trying to reverse engineer the values of an unclear parameter by comparing R outcomes to Stata outcomes for an ordinal probit model (link). To my surprise I could not replicate the outcome (fake data set with 2000 observations). To be sure, I decided to replicate the ordinal probit models with a Stata dataset. As the results (below) show, the differences in output were not a coincidence. I provided all data and code for both Stata and R below Although I did not expect the results to be completely identical, they are off by quite some margin.

Because of this, I have two questions:

1. Is there an R package that is able to replicate the Stata results, so that I can reverse engineer the values that I am looking for?
2. Can anyone shed any sort of light on these differences. What is it about these models that makes these differences so large and what does that say about any result from an ordinal model?

References: https://www.stata.com/manuals14/roprobit.pdf

NOTE: Everyone, please feel encouraged to check if there are any mistakes in my code that might explain the differences

# Replication

In Stata, with oprobit:

use http://www.stata-press.com/data/r14/fullauto

oprobit rep77 foreign length mpg

Iteration 0:   log likelihood = -89.895098
Iteration 1:   log likelihood = -78.106316
Iteration 2:   log likelihood = -78.020086
Iteration 3:   log likelihood = -78.020025
Iteration 4:   log likelihood = -78.020025

Ordered probit regression                       Number of obs     =         66
LR chi2(3)        =      23.75
Prob > chi2       =     0.0000
Log likelihood = -78.020025                     Pseudo R2         =     0.1321

------------------------------------------------------------------------------
rep77 |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
foreign |   1.704861   .4246796     4.01   0.000     .8725037    2.537217
length |   .0468675    .012648     3.71   0.000      .022078    .0716571
mpg |   .1304559   .0378628     3.45   0.001     .0562463    .2046656
-------------+----------------------------------------------------------------
/cut1 |    10.1589   3.076754                      4.128577    16.18923
/cut2 |   11.21003   3.107527                      5.119389    17.30067
/cut3 |   12.54561   3.155233                      6.361467    18.72975
/cut4 |   13.98059   3.218793                      7.671874    20.28931
------------------------------------------------------------------------------


In R, with polr from the MASS package:

library(MASS)
polr  <- polr(as.ordered(rep77) ~ foreign + length + mpg, Hess=TRUE, data=fullauto);summary(polr)

Call:
polr(formula = as.ordered(rep77) ~ foreign + length + mpg, data = fullauto,
Hess = TRUE)

Coefficients:
Value Std. Error t value
foreign 2.89680    0.79106   3.662
length  0.08283    0.02277   3.637
mpg     0.23077    0.07053   3.272

Intercepts:
Value   Std. Error t value
1|2 17.9273  5.5637     3.2222
2|3 19.8649  5.6071     3.5428
3|4 22.1031  5.7188     3.8650
4|5 24.6919  5.9037     4.1825

Residual Deviance: 156.5014
AIC: 170.5014


In R, with vglm from the VGAM package:

library(VGAM)
vglm <- vglm(as.ordered(rep77) ~ foreign + length + mpg, family=propodds, data=fullauto);summary(vglm)

Call:
vglm(formula = as.ordered(rep77) ~ foreign + length + mpg, family = propodds,
data = fullauto)

Coefficients:
Estimate Std. Error z value Pr(>|z|)
(Intercept):1 -17.92738    5.47221  -3.276 0.001053 **
(Intercept):2 -19.86496    5.52892  -3.593 0.000327 ***
(Intercept):3 -22.10321    5.63780  -3.921 8.84e-05 ***
(Intercept):4 -24.69203    5.81757  -4.244 2.19e-05 ***
foreign         2.89684    0.75766   3.823 0.000132 ***
length          0.08283    0.02252   3.677 0.000236 ***
mpg             0.23076    0.06805   3.391 0.000696 ***
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Residual deviance: 156.5014 on 257 degrees of freedom

Log-likelihood: -78.2507 on 257 degrees of freedom

Number of Fisher scoring iterations: 6

No Hauck-Donner effect found in any of the estimates

Exponentiated coefficients:
foreign    length       mpg
18.116799  1.086354  1.259563


In R, with clm from the ordinal package:

clm <- clm(as.ordered(rep77) ~ foreign + length + mpg, data=fullauto)
summary(clm)

formula: as.ordered(rep77) ~ foreign + length + mpg
data:    fullauto

logit flexible  66   -78.25 170.50 6(0)  5.65e-10 8.9e+07

Coefficients:
Estimate Std. Error z value Pr(>|z|)
foreign  2.89681    0.79064   3.664 0.000248 ***
length   0.08283    0.02272   3.646 0.000267 ***
mpg      0.23077    0.07045   3.275 0.001055 **
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Threshold coefficients:
Estimate Std. Error z value
1|2   17.927      5.551   3.229
2|3   19.865      5.596   3.550
3|4   22.103      5.709   3.872
4|5   24.692      5.891   4.192
(8 observations deleted due to missingness)


In R, with orm from the rms package:

library(rms)
orm <- orm(as.ordered(rep77) ~ foreign + length + mpg, family=probit, data=fullauto)
summary(orm)

Effects              Response : as.ordered(rep77)

Factor  Low High   Diff. Effect  S.E.    Lower 0.95 Upper 0.95
foreign   0   1.00  1.00 1.70480 0.42468 0.87249    2.5372
length  170 203.75 33.75 1.58180 0.42687 0.74512    2.4184
mpg      18  24.75  6.75 0.88057 0.25557 0.37966    1.3815


# DATA for R

fullauto <- structure(list(make = structure(c(1, 1, 1, 2, 2, 3, 4, 4, 4,
4, 4, 4, 4, 5, 5, 5, 6, 6, 6, 6, 6, 6, 7, 7, 7, 7, 8, 8, 8, 8,
9, 10, 10, 11, 11, 12, 12, 12, 13, 14, 14, 14, 14, 14, 14, 15,
15, 15, 15, 15, 15, 15, 16, 17, 17, 17, 17, 17, 18, 18, 18, 18,
18, 18, 19, 20, 21, 21, 21, 22, 22, 22, 22, 23), label = "Make", format.stata = "%8.0g", class = c("haven_labelled",
"vctrs_vctr", "double"), labels = c(AMC = 1, Audi = 2, BMW = 3,
Buick = 4, Cad. = 5, Chev. = 6, Datsun = 7, Dodge = 8, Fiat = 9,
Ford = 10, Honda = 11, Linc. = 12, Mazda = 13, Merc. = 14, Olds = 15,
Peugeot = 16, Plym. = 17, Pont. = 18, Renault = 19, Subaru = 20,
Toyota = 21, VW = 22, Volvo = 23)), model = structure(c(1, 2,
3, 4, 5000, 320, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17,
18, 19, 20, 200, 210, 510, 810, 21, 22, 23, 24, 25, 26, 27, 28,
29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44,
45, 98, 604, 46, 47, 48, 49, 50, 51, 52, 53, 54, 55, 56, 57,
58, 59, 60, 61, 62, 63, 64, 65, 260), label = "Model", format.stata = "%8.0g", class = c("haven_labelled",
"vctrs_vctr", "double"), labels = c(Concord = 1, Pacer = 2, Spirit = 3,
Fox = 4, Century = 5, Electra = 6, LeSabre = 7, Opel = 8, Regal = 9,
Riviera = 10, Skylark = 11, Deville = 12, Eldrado = 13, Seville = 14,
Chevette = 15, Impala = 16, Malibu = 17, MCarlo = 18, Monza = 19,
Nova = 20, Colt = 21, Diplomat = 22, Magnum = 23, StRegis = 24,
Strada = 25, Fiesta = 26, Mustang = 27, Accord = 28, Civic = 29,
Cntntl = 30, Mark V = 31, Vrsills = 32, GLC = 33, Bobcat = 34,
Cougar = 35, XR-7 = 36, Marquis = 37, Monarch = 38, Zephyr = 39,
Cutlass = 40, CutlSupr = 41, Delta 88 = 42, Omega = 43, Starfire = 44,
Toronado = 45, Arrow = 46, Champ = 47, Horizon = 48, Sapporo = 49,
Volare = 50, Catalina = 51, Firebird = 52, GranPrix = 53, Le Mans = 54,
Phoenix = 55, Sunbird = 56, Le Car = 57, Subaru = 58, Celica = 59,
Corolla = 60, Corona = 61, Rabbit = 62, Diesel = 63, Scirocco = 64,
Dasher = 65)), price = structure(c(4099, 4749, 3799, 6295, 9690,
9735, 4816, 7827, 5788, 4453, 5189, 10372, 4082, 11385, 14500,
15906, 3299, 5705, 4504, 5104, 3667, 3955, 6229, 4589, 5079,
8129, 3984, 4010, 5886, 6342, 4296, 4389, 4187, 5799, 4499, 11497,
13594, 13466, 3995, 3829, 5379, 6303, 6165, 4516, 3291, 4733,
5172, 4890, 4181, 4195, 10371, 8814, 12990, 4647, 4425, 4482,
6486, 4060, 5798, 4934, 5222, 4723, 4424, 4172, 3895, 3798, 5899,
3748, 5719, 4697, 5397, 6850, 7140, 11995), label = "Price", format.stata = "%8.0g"),
mpg = structure(c(22, 17, 22, 23, 17, 25, 20, 15, 18, 26,
20, 16, 19, 14, 14, 21, 29, 16, 22, 22, 24, 19, 23, 35, 24,
21, 30, 18, 16, 17, 21, 28, 21, 25, 28, 12, 12, 14, 30, 22,
14, 14, 15, 18, 20, 19, 19, 18, 19, 24, 16, 21, 14, 38, 34,
25, 26, 18, 18, 18, 19, 19, 19, 24, 26, 35, 18, 31, 18, 25,
41, 25, 23, 17), label = "Mileage (mpg)", format.stata = "%8.0g"),
rep78 = structure(c(3, 3, NA, 3, 5, 4, 3, 4, 3, NA, 3, 3,
3, 3, 2, 3, 3, 4, 3, 2, 2, 3, 4, 5, 4, 4, 5, 2, 2, 2, 3,
4, 3, 5, 4, 3, 3, 3, 4, 4, 4, 4, 3, 3, 3, 3, 3, 4, 3, 1,
3, 4, NA, 3, 5, 3, NA, 2, 4, 1, 3, 3, NA, 2, 3, 5, 5, 5,
5, 4, 5, 4, 4, 5), label = "Repair Record 1978", format.stata = "%9.0g", class = c("haven_labelled",
"vctrs_vctr", "double"), labels = c(Poor = 1, Fair = 2, Average = 3,
Good = 4, Excellent = 5)), rep77 = structure(c(2, 1, NA,
3, 2, 4, 3, 4, 4, NA, 3, 4, 3, 3, 2, 3, 3, 4, 3, 3, 2, 3,
3, 5, 4, 4, 4, 2, 2, 2, 1, NA, 3, 5, 4, 4, 4, 3, 4, 3, 3,
4, 2, NA, 3, 3, 4, 4, 3, 1, 3, 4, NA, 3, 4, NA, NA, 2, 4,
2, 3, 3, NA, 2, 3, 4, 5, 5, 5, 3, 4, 3, 3, 3), label = "Repair Record 1977", format.stata = "%9.0g", class = c("haven_labelled",
"vctrs_vctr", "double"), labels = c(Poor = 1, Fair = 2, Average = 3,
Good = 4, Excellent = 5)), hdroom = structure(c(2.5, 3, 3,
2.5, 3, 2.5, 4.5, 4, 4, 3, 2, 3.5, 3.5, 4, 3.5, 3, 2.5, 4,
3.5, 2, 2, 3.5, 1.5, 2, 2.5, 2.5, 2, 4, 4, 4.5, 2.5, 1.5,
2, 3, 2.5, 3.5, 2.5, 3.5, 3.5, 3, 3.5, 3, 3.5, 3, 3.5, 4.5,
2, 4, 4.5, 2, 3.5, 4, 3.5, 2, 2.5, 4, 1.5, 5, 4, 1.5, 2,
3.5, 3.5, 2, 3, 2.5, 2.5, 3, 2, 3, 3, 2, 2.5, 2.5), label = "Headroom (in.)", format.stata = "%6.1f"),
rseat = structure(c(27.5, 25.5, 18.5, 28, 27, 26, 29, 31.5,
30.5, 24, 28.5, 30, 27, 31.5, 30, 30, 26, 29.5, 28.5, 28.5,
25, 27, 21, 23.5, 22, 27, 24, 29, 29, 28, 26.5, 26, 23, 25.5,
23.5, 30.5, 28.5, 27, 25.5, 25.5, 29.5, 25, 30.5, 27, 29,
28, 28, 29, 27, 25.5, 30, 31.5, 30.5, 21.5, 23, 25, 22, 31,
29, 23.5, 28.5, 28, 27, 25, 23, 25.5, 22, 24.5, 23, 25.5,
25.5, 23.5, 37.5, 29.5), label = "Rear Seat (in.)", format.stata = "%6.1f"),
trunk = structure(c(11, 11, 12, 11, 15, 12, 16, 20, 21, 10,
16, 17, 13, 20, 16, 13, 9, 20, 17, 16, 7, 13, 6, 8, 8, 8,
8, 17, 17, 21, 16, 9, 10, 10, 5, 22, 18, 15, 11, 9, 16, 16,
23, 15, 17, 16, 16, 20, 14, 10, 17, 20, 14, 11, 11, 17, 8,
16, 20, 7, 16, 17, 13, 7, 10, 11, 14, 9, 11, 15, 15, 16,
12, 14), label = "Trunk space (cu. ft.)", format.stata = "%8.0g"),
weight = structure(c(2930, 3350, 2640, 2070, 2830, 2650,
3250, 4080, 3670, 2230, 3280, 3880, 3400, 4330, 3900, 4290,
2110, 3690, 3180, 3220, 2750, 3430, 2370, 2020, 2280, 2750,
2120, 3600, 3600, 3740, 2130, 1800, 2650, 2240, 1760, 4840,
4720, 3830, 1980, 2580, 4060, 4130, 3720, 3370, 2830, 3300,
3310, 3690, 3370, 2730, 4030, 4060, 3420, 3260, 1800, 2200,
2520, 3330, 3700, 3470, 3210, 3200, 3420, 2690, 1830, 2050,
2410, 2200, 2670, 1930, 2040, 1990, 2160, 3170), label = "Weight (lbs.)", format.stata = "%8.0g"),
length = structure(c(186, 173, 168, 174, 189, 177, 196, 222,
218, 170, 200, 207, 200, 221, 204, 204, 163, 212, 193, 200,
179, 197, 170, 165, 170, 184, 163, 206, 206, 220, 161, 147,
179, 172, 149, 233, 230, 201, 154, 169, 221, 217, 212, 198,
195, 198, 198, 218, 200, 180, 206, 220, 192, 170, 157, 165,
182, 201, 214, 198, 201, 199, 203, 179, 142, 164, 174, 165,
175, 155, 155, 156, 172, 193), label = "Length (in.)", format.stata = "%8.0g"),
turn = structure(c(40, 40, 35, 36, 37, 34, 40, 43, 43, 34,
42, 43, 42, 44, 43, 45, 34, 43, 31, 41, 40, 43, 35, 32, 34,
38, 35, 46, 46, 46, 36, 33, 43, 36, 34, 51, 48, 41, 33, 39,
48, 45, 44, 41, 43, 42, 42, 42, 43, 40, 43, 43, 38, 37, 37,
36, 38, 44, 42, 42, 45, 40, 43, 41, 34, 36, 36, 35, 36, 35,
35, 36, 36, 37), label = "Turn Circle (ft.) ", format.stata = "%8.0g"),
displ = structure(c(121, 258, 121, 97, 131, 121, 196, 350,
231, 304, 196, 231, 231, 425, 350, 350, 231, 250, 200, 200,
151, 250, 119, 85, 119, 146, 98, 318, 318, 225, 105, 98,
140, 107, 91, 400, 400, 302, 86, 140, 302, 302, 302, 250,
140, 231, 231, 231, 231, 151, 350, 350, 163, 156, 86, 105,
119, 225, 231, 231, 231, 231, 231, 151, 79, 97, 134, 97,
134, 89, 90, 97, 97, 163), label = "Displacement (cu. in.)", format.stata = "%8.0g"),
gratio = structure(c(3.57999992370605, 2.52999997138977,
3.07999992370605, 3.70000004768372, 3.20000004768372, 3.64000010490417,
2.9300000667572, 2.41000008583069, 2.73000001907349, 2.86999988555908,
2.9300000667572, 2.9300000667572, 3.07999992370605, 2.27999997138977,
2.19000005722046, 2.24000000953674, 2.9300000667572, 2.55999994277954,
2.73000001907349, 2.73000001907349, 2.73000001907349, 2.55999994277954,
3.89000010490417, 3.70000004768372, 3.53999996185303, 3.54999995231628,
3.53999996185303, 2.47000002861023, 2.47000002861023, 2.94000005722046,
3.36999988555908, 3.15000009536743, 3.07999992370605, 3.04999995231628,
3.29999995231628, 2.47000002861023, 2.47000002861023, 2.47000002861023,
3.73000001907349, 2.73000001907349, 2.75, 2.75, 2.25999999046326,
2.4300000667572, 3.07999992370605, 2.9300000667572, 2.9300000667572,
2.73000001907349, 3.07999992370605, 2.73000001907349, 2.41000008583069,
2.41000008583069, 3.57999992370605, 3.04999995231628, 2.97000002861023,
3.36999988555908, 3.53999996185303, 3.23000001907349, 2.73000001907349,
3.07999992370605, 2.9300000667572, 2.9300000667572, 3.07999992370605,
2.73000001907349, 3.72000002861023, 3.80999994277954, 3.05999994277954,
3.21000003814697, 3.04999995231628, 3.77999997138977, 3.77999997138977,
3.77999997138977, 3.74000000953674, 2.98000001907349), label = "Gear Ratio", format.stata = "%6.2f"),
order = structure(c(1, 2, 3, 5, 4, 6, 7, 8, 9, 10, 11, 12,
13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27,
28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42,
43, 44, 45, 47, 48, 49, 50, 51, 52, 46, 53, 54, 55, 56, 57,
58, 59, 60, 61, 62, 63, 64, 65, 66, 67, 68, 69, 70, 71, 72,
73, 74), label = "Original order", format.stata = "%8.0g"),
foreign = structure(c(0, 0, 0, 1, 1, 1, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0, 1,
0, 0, 1, 1, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1), label = "Foreign", format.stata = "%8.0g", class = c("haven_labelled",
"vctrs_vctr", "double"), labels = c(Domestic = 0, Foreign = 1
)), wgtd = structure(c(2930, 3350, 2640, NA, NA, NA, 3250,
4080, 3670, 2230, 3280, 3880, 3400, 4330, 3900, 4290, 2110,
3690, 3180, 3220, 2750, 3430, NA, NA, NA, NA, 2120, 3600,
3600, 3740, NA, 1800, 2650, NA, NA, 4840, 4720, 3830, NA,
2580, 4060, 4130, 3720, 3370, 2830, 3300, 3310, 3690, 3370,
2730, 4030, 4060, NA, 3260, 1800, 2200, 2520, 3330, 3700,
3470, 3210, 3200, 3420, 2690, NA, NA, NA, NA, NA, NA, NA,
NA, NA, NA), format.stata = "%9.0g"), wgtf = structure(c(NA,
NA, NA, 2070, 2830, 2650, NA, NA, NA, NA, NA, NA, NA, NA,
NA, NA, NA, NA, NA, NA, NA, NA, 2370, 2020, 2280, 2750, NA,
NA, NA, NA, 2130, NA, NA, 2240, 1760, NA, NA, NA, 1980, NA,
NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, 3420, NA,
NA, NA, NA, NA, NA, NA, NA, NA, NA, NA, 1830, 2050, 2410,
2200, 2670, 1930, 2040, 1990, 2160, 3170), format.stata = "%9.0g")), label = "Automobile Models", row.names = c(NA,
-74L), class = c("tbl_df", "tbl", "data.frame"))

• I don't think polr(·) defaults to probit...try including the argument method="probit" Apr 29, 2021 at 14:05
• Thank you, that is very helpful. Will update!
– Tom
Apr 29, 2021 at 14:07
• @GreggH That was it.. I cannot believe I missed that.. Thank you so much!
– Tom
Apr 29, 2021 at 14:09
• I guess that means I can delete the question.
– Tom
Apr 29, 2021 at 14:09
• I'd leave it up...other's may encounter a similar problem Apr 29, 2021 at 15:26

As per Gregg H's suggestions, I will leave up this post (and put an answer, just to show that this post has a solution).

The problem was that in polr, (as well as vglm and clm), the probit method was not explicitly specified.

polr  <- polr(as.ordered(rep77) ~ foreign + length + mpg, method="probit", Hess=TRUE,
data=fullauto);summary(polr)

Call:
polr(formula = as.ordered(rep77) ~ foreign + length + mpg, data = fullauto,
Hess = TRUE, method = "probit")

Coefficients:
Value Std. Error t value
foreign 1.70386    0.42478   4.011
length  0.04681    0.01264   3.702
mpg     0.13032    0.03786   3.442

Intercepts:
Value   Std. Error t value
1|2 10.1452  3.0762     3.2980
2|3 11.1963  3.1068     3.6038
3|4 12.5316  3.1544     3.9727
4|5 13.9663  3.2185     4.3394

Residual Deviance: 156.0401
AIC: 170.0401
(8 observations deleted due to missingness)


Stata:

use http://www.stata-press.com/data/r14/fullauto

oprobit rep77 foreign length mpg

Iteration 0:   log likelihood = -89.895098
Iteration 1:   log likelihood = -78.106316
Iteration 2:   log likelihood = -78.020086
Iteration 3:   log likelihood = -78.020025
Iteration 4:   log likelihood = -78.020025

Ordered probit regression                       Number of obs     =         66
LR chi2(3)        =      23.75
Prob > chi2       =     0.0000
Log likelihood = -78.020025                     Pseudo R2         =     0.1321

------------------------------------------------------------------------------
rep77 |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
foreign |   1.704861   .4246796     4.01   0.000     .8725037    2.537217
length |   .0468675    .012648     3.71   0.000      .022078    .0716571
mpg |   .1304559   .0378628     3.45   0.001     .0562463    .2046656
-------------+----------------------------------------------------------------
/cut1 |    10.1589   3.076754                      4.128577    16.18923
/cut2 |   11.21003   3.107527                      5.119389    17.30067
/cut3 |   12.54561   3.155233                      6.361467    18.72975
/cut4 |   13.98059   3.218793                      7.671874    20.28931
------------------------------------------------------------------------------