List,
I'm working on fitting a logistic model for a well known dataset (which is
given below in case anyone wants to try to reproduce). I used both R and
SAS to fit the model and have some differences in the parameter estimates.
I'm wondering if R calculates the ML estimates differently. I'm making
NO
accusations as to which program is "right or wrong". That is not the
focus
of this posting. As a "newer" R user I'm trying to understand the
algorithm
that R might use to calculate ML estimation. The largest difference seems
to with the race factors. R gives a p-value of 0.46995 for race=black and
SAS gives a p-value of 0.0753 for race=black. Clearly one is borderline
significant and the other is not. Many thanks to all who might be able to
offer any insight on this. Both R and SAS code and output are included in
this message (along with the dataset).
Thanks,
Patrick
MY R CODE IS:
Dataset <- read.table("<path>", header=TRUE,
sep="", na.strings="NA",
dec=".", strip.white=TRUE)
Dataset$race <- factor(Dataset$race,
levels=c('other','black','white'))
GLM.1 <- glm(low ~ lwt + ptl + ht + race + smoke ,
family=binomial(logit), data=Dataset)
summary(GLM.1)
MY SAS CODE IS:
PROC LOGISTIC descending DATA=p2;
class race (ref='other');
MODEL LOW = lwt ptl ht race smoke / lackfit parmlabel expb link=logit;
RUN;
MY R OUTPUT IS:
Coefficients:
Estimate Std. Error z value Pr(>|z|)
(Intercept) 0.92619 0.85549 1.083 0.27897
lwt -0.01650 0.00692 -2.384 0.01712 *
ptl 1.23116 0.44607 2.760 0.00578 **
ht 1.76197 0.70748 2.490 0.01276 *
race[T.black] 0.39552 0.54739 0.723 0.46995
race[T.white] -0.86291 0.43517 -1.983 0.04737 *
smoke 0.88007 0.40049 2.197 0.02798 *
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05
'.' 0.1 ' ' 1
(Dispersion parameter for binomial family taken to be 1)
Null deviance: 234.67 on 188 degrees of freedom
Residual deviance: 200.62 on 182 degrees of freedom
AIC: 214.62
Number of Fisher Scoring iterations: 4
MY SAS OUTPUT IS:
The LOGISTIC Procedure
Analysis of Maximum Likelihood Estimates
Standard Wald
Parameter DF Estimate Error Chi-Square Pr > ChiSq Exp(Est)
Label
Intercept 1 0.9287 0.9326 0.9916 0.3193 2.531
Intercept: low=1
lwt 1 -0.0173 0.00699 6.1425 0.0132 0.983
ptl 1 1.1958 0.4472 7.1493 0.0075 3.306
ht 1 1.7482 0.7090 6.0805 0.0137 5.745
race black 1 0.5963 0.3352 3.1643 0.0753 1.815
race black
race white 1 -0.7200 0.2668 7.2803 0.0070 0.487
race white
smoke 1 0.8648 0.4009 4.6534 0.0310 2.375
0 19 182 black 0 0 1 0 0 2523
0 33 155 other 0 0 0 1 0 2551
0 20 105 white 1 0 0 1 0 2557
0 21 108 white 1 0 1 1 0 2594
0 18 107 white 1 0 1 0 0 2600
0 21 124 other 0 0 0 0 0 2622
0 22 118 white 0 0 0 1 0 2637
0 17 103 other 0 0 0 1 0 2637
0 29 123 white 1 0 0 1 0 2663
0 26 113 white 1 0 0 0 0 2665
0 19 95 other 0 0 0 0 0 2722
0 19 150 other 0 0 0 1 0 2733
0 22 95 other 0 1 0 0 0 2750
0 30 107 other 0 0 1 1 1 2750
0 18 100 white 1 0 0 0 0 2769
0 15 98 black 0 0 0 0 0 2778
0 25 118 white 1 0 0 1 0 2782
0 20 120 other 0 0 1 0 0 2807
0 28 120 white 1 0 0 1 0 2821
0 32 101 other 0 0 0 1 0 2835
0 31 100 white 0 0 1 1 0 2835
0 36 202 white 0 0 0 1 0 2836
0 28 120 other 0 0 0 0 0 2863
0 25 120 other 0 0 1 1 0 2877
0 28 167 white 0 0 0 0 0 2877
0 17 122 white 1 0 0 0 0 2906
0 29 150 white 0 0 0 1 0 2920
0 26 168 black 1 0 0 0 0 2920
0 17 113 black 0 0 0 1 0 2920
0 24 90 white 1 0 0 1 1 2948
0 35 121 black 1 0 0 1 1 2948
0 25 155 white 0 0 0 1 0 2977
0 25 125 black 0 0 0 0 0 2977
0 29 140 white 1 0 0 1 0 2977
0 19 138 white 1 0 0 1 0 2977
0 27 124 white 1 0 0 0 0 2992
0 31 215 white 1 0 0 1 0 3005
0 33 109 white 1 0 0 1 0 3033
0 21 185 black 1 0 0 1 0 3042
0 19 189 white 0 0 0 1 0 3062
0 23 130 black 0 0 0 1 0 3062
0 21 160 white 0 0 0 0 0 3062
0 18 90 white 1 0 1 0 0 3076
0 18 90 white 1 0 1 0 0 3076
0 32 132 white 0 0 0 1 0 3080
0 19 132 other 0 0 0 0 0 3090
0 24 115 white 0 0 0 1 0 3090
0 22 85 other 1 0 0 0 0 3090
0 22 120 white 0 1 0 1 0 3100
0 23 128 other 0 0 0 0 0 3104
0 22 130 white 1 0 0 0 0 3132
0 30 95 white 1 0 0 1 0 3147
0 19 115 other 0 0 0 0 0 3175
0 16 110 other 0 0 0 0 0 3175
0 21 110 other 1 0 1 0 0 3203
0 30 153 other 0 0 0 0 0 3203
0 20 103 other 0 0 0 0 0 3203
0 17 119 other 0 0 0 0 0 3225
0 23 119 other 0 0 0 1 0 3232
0 24 110 other 0 0 0 0 0 3232
0 28 140 white 0 0 0 0 0 3234
0 26 133 other 1 0 0 0 1 3260
0 20 169 other 0 0 1 1 1 3274
0 24 115 other 0 0 0 1 0 3274
0 28 250 other 1 0 0 1 0 3303
0 20 141 white 0 0 1 1 1 3317
0 22 158 black 0 0 0 1 1 3317
0 22 112 white 1 0 0 0 1 3317
0 31 150 other 1 0 0 1 0 3321
0 23 115 other 1 0 0 1 0 3331
0 16 112 black 0 0 0 0 0 3374
0 16 135 white 1 0 0 0 0 3374
0 18 229 black 0 0 0 0 0 3402
0 25 140 white 0 0 0 1 0 3416
0 32 134 white 1 0 0 1 1 3430
0 20 121 black 1 0 0 0 0 3444
0 23 190 white 0 0 0 0 0 3459
0 22 131 white 0 0 0 1 0 3460
0 32 170 white 0 0 0 0 0 3473
0 30 110 other 0 0 0 0 0 3475
0 20 127 other 0 0 0 0 0 3487
0 23 123 other 0 0 0 0 0 3544
0 17 120 other 1 0 0 0 0 3572
0 19 105 other 0 0 0 0 0 3572
0 23 130 white 0 0 0 0 0 3586
0 36 175 white 0 0 0 0 0 3600
0 22 125 white 0 0 0 1 0 3614
0 24 133 white 0 0 0 0 0 3614
0 21 134 other 0 0 0 1 0 3629
0 19 235 white 1 1 0 0 0 3629
0 25 95 white 1 0 1 0 1 3637
0 16 135 white 1 0 0 0 0 3643
0 29 135 white 0 0 0 1 0 3651
0 29 154 white 0 0 0 1 0 3651
0 19 147 white 1 0 0 0 0 3651
0 19 147 white 1 0 0 0 0 3651
0 30 137 white 0 0 0 1 0 3699
0 24 110 white 0 0 0 1 0 3728
0 19 184 white 1 1 0 0 0 3756
0 24 110 other 0 0 0 0 1 3770
0 23 110 white 0 0 0 1 0 3770
0 20 120 other 0 0 0 0 0 3770
0 25 241 black 0 1 0 0 0 3790
0 30 112 white 0 0 0 1 0 3799
0 22 169 white 0 0 0 0 0 3827
0 18 120 white 1 0 0 1 0 3856
0 16 170 black 0 0 0 1 0 3860
0 32 186 white 0 0 0 1 0 3860
0 18 120 other 0 0 0 1 0 3884
0 29 130 white 1 0 0 1 0 3884
0 33 117 white 0 0 1 1 0 3912
0 20 170 white 1 0 0 0 0 3940
0 28 134 other 0 0 0 1 0 3941
0 14 135 white 0 0 0 0 0 3941
0 28 130 other 0 0 0 0 0 3969
0 25 120 white 0 0 0 1 0 3983
0 16 95 other 0 0 0 1 0 3997
0 20 158 white 0 0 0 1 0 3997
0 26 160 other 0 0 0 0 0 4054
0 21 115 white 0 0 0 1 0 4054
0 22 129 white 0 0 0 0 0 4111
0 25 130 white 0 0 0 1 0 4153
0 31 120 white 0 0 0 1 0 4167
0 35 170 white 0 0 0 1 1 4174
0 19 120 white 1 0 0 0 0 4238
0 24 116 white 0 0 0 1 0 4593
0 45 123 white 0 0 0 1 0 4990
1 28 120 other 1 0 1 0 1 709
1 29 130 white 0 0 1 1 0 1021
1 34 187 black 1 1 0 0 0 1135
1 25 105 other 0 1 0 0 1 1330
1 25 85 other 0 0 1 0 0 1474
1 27 150 other 0 0 0 0 0 1588
1 23 97 other 0 0 1 1 0 1588
1 24 124 black 0 0 0 1 1 1701
1 24 132 other 0 1 0 0 0 1729
1 21 165 white 1 1 0 1 0 1790
1 32 105 white 1 0 0 0 0 1818
1 19 91 white 1 0 1 0 1 1885
1 25 115 other 0 0 0 0 0 1893
1 16 130 other 0 0 0 1 0 1899
1 25 92 white 1 0 0 0 0 1928
1 20 150 white 1 0 0 1 0 1928
1 21 200 black 0 0 1 1 0 1928
1 24 155 white 1 0 0 0 1 1936
1 21 103 other 0 0 0 0 0 1970
1 20 125 other 0 0 1 0 0 2055
1 25 89 other 0 0 0 1 1 2055
1 19 102 white 0 0 0 1 0 2082
1 19 112 white 1 0 1 0 0 2084
1 26 117 white 1 0 0 0 1 2084
1 24 138 white 0 0 0 0 0 2100
1 17 130 other 1 0 1 0 1 2125
1 20 120 black 1 0 0 1 0 2126
1 22 130 white 1 0 1 1 1 2187
1 27 130 black 0 0 1 0 0 2187
1 20 80 other 1 0 1 0 0 2211
1 17 110 white 1 0 0 0 0 2225
1 25 105 other 0 0 0 1 1 2240
1 20 109 other 0 0 0 0 0 2240
1 18 148 other 0 0 0 0 0 2282
1 18 110 black 1 0 0 0 1 2296
1 20 121 white 1 0 1 0 1 2296
1 21 100 other 0 0 0 1 1 2301
1 26 96 other 0 0 0 0 0 2325
1 31 102 white 1 0 0 1 1 2353
1 15 110 white 0 0 0 0 0 2353
1 23 187 black 1 0 0 1 0 2367
1 20 122 black 1 0 0 0 0 2381
1 24 105 black 1 0 0 0 0 2381
1 15 115 other 0 0 1 0 0 2381
1 23 120 other 0 0 0 0 0 2395
1 30 142 white 1 0 0 0 1 2410
1 22 130 white 1 0 0 1 0 2410
1 17 120 white 1 0 0 1 0 2414
1 23 110 white 1 0 0 0 1 2424
1 17 120 black 0 0 0 1 0 2438
1 26 154 other 0 1 0 1 1 2442
1 20 105 other 0 0 0 1 0 2450
1 26 190 white 1 0 0 0 0 2466
1 14 101 other 1 0 0 0 1 2466
1 28 95 white 1 0 0 1 0 2466
1 14 100 other 0 0 0 1 0 2495
1 23 94 other 1 0 0 0 0 2495
1 17 142 black 0 1 0 0 0 2495
1 21 130 white 1 1 0 1 0 2495
On 11/03/2008, at 6:09 AM, Patrick Richardson wrote:> List, > > I'm working on fitting a logistic model for a well known dataset > (which is > given below in case anyone wants to try to reproduce). I used both > R and > SAS to fit the model and have some differences in the parameter > estimates. > I'm wondering if R calculates the ML estimates differently. I'm > making NO > accusations as to which program is "right or wrong". That is not > the focus > of this posting. As a "newer" R user I'm trying to understand the > algorithm > that R might use to calculate ML estimation. The largest > difference seems > to with the race factors. R gives a p-value of 0.46995 for > race=black and > SAS gives a p-value of 0.0753 for race=black. Clearly one is > borderline > significant and the other is not. Many thanks to all who might be > able to > offer any insight on this. Both R and SAS code and output are > included in > this message (along with the dataset).Try setting options(contrasts=c("contr.SAS","contr.poly")) before you run your analysis in R. cheers, Rolf Turner ###################################################################### Attention:\ This e-mail message is privileged and confid...{{dropped:9}}
Apparently Analagous Threads
- Stochastic Gradient Ascent for logistic regression
- How to formulate an (effect-modifying) interaction with matching variable in a conditional logistic regression?
- Logistic regression: At times correlation matrix of coefficients gets messed up
- Re : Seperate timestamp data into date and time
- MCMCpack: Selecting a better model using BayesFactor