Logistic Regression/ Hosmer test

eilayyan · Posted 02-18-2012 02:27 PM

Hello,

I run a logistic regression in SAS with this code:

proc logistic data = model3;

class gender(param = ref ref='0')smoking(param = ref ref='0')_0SEC6 (param = ref ref='0');

model ACT5c (event = '1') = age gender smoking fev_con bmi re_norm PF_Norm EF_Norm sf_norm _0SEC6/ rsq lackfit;

run;

the p value in hosmer test is 0.655

But when i changed the event from 1 to 0 in this code:

proc logistic data = model3;

class gender(param = ref ref='0')smoking(param = ref ref='0')_0SEC6 (param = ref ref='0');

model ACT5c (event = '0') = age gender smoking fev_con bmi re_norm PF_Norm EF_Norm sf_norm _0SEC6/ rsq lackfit;

run;

the p value in hosmer test is 0.9

so how that come? it is same data and same code!!!

Could you please help me in that.

Thanks

Owis

art297 · Posted 02-18-2012 05:54 PM

I am NOT a statistician, and don't know why you would obtain those differences, but I used the data from one of the examples from the documentation and was able to replicate your findings. The first example doesn't show the effect. In the second example, I changed all of the pain values to equal 0 except for the last column.

I was able to get different hosmer values if there were only 1 one for pain for each age group.

Hopefully, that might give you a clue as to what to look for.

Data Neuralgia;

input Treatment $ Sex $ Age Duration Pain $ @@;

datalines;

1 0 68 1 0 1 1 74 16 0 1 0 67 30 0

1 1 66 26 1 1 0 67 28 0 1 0 77 16 0

0 0 71 12 0 1 0 72 50 0 1 0 76 9 1

0 1 71 17 1 0 0 63 27 0 0 0 69 18 1

1 0 66 12 0 0 1 62 42 0 1 0 64 1 1

0 0 64 17 0 1 1 74 4 0 0 0 72 25 0

1 1 70 1 1 1 1 66 19 0 1 1 59 29 0

0 0 64 30 0 0 1 70 28 0 0 1 69 1 0

1 0 78 1 0 1 1 83 1 1 1 0 69 42 0

1 1 75 30 1 1 1 77 29 1 1 0 79 20 1

0 1 70 12 0 0 0 69 12 0 1 0 65 14 0

1 1 70 1 0 1 1 67 23 0 0 1 76 25 1

1 1 78 12 1 1 1 77 1 1 1 0 69 24 0

1 1 66 4 1 1 0 65 29 0 1 1 60 26 1

0 1 78 15 1 1 1 75 21 1 0 0 67 11 0

1 0 72 27 0 1 0 70 13 1 0 1 75 6 1

1 0 65 7 0 1 0 68 27 1 1 1 68 11 1

1 1 67 17 1 1 1 70 22 0 0 1 65 15 0

1 0 67 1 1 0 1 67 10 0 1 0 72 11 1

0 0 74 1 0 1 1 80 21 1 0 0 69 3 0

;

proc logistic data=Neuralgia;

class treatment(param = ref ref='0')sex(param = ref ref='0');

model pain (event = '1') = Treatment Sex Age Duration/ rsq lackfit;

run;

proc logistic data=Neuralgia;

class treatment(param = ref ref='0')sex(param = ref ref='0');

model pain (event = '0') = Treatment Sex Age Duration/ rsq lackfit;

run;

Data Neuralgia;

input Treatment $ Sex $ Age Duration Pain $ @@;

datalines;

1 0 68 1 0 1 1 74 16 0 1 0 67 30 0

1 1 66 26 0 1 0 67 28 0 1 0 77 16 0

0 0 71 12 0 1 0 72 50 0 1 0 76 9 0

0 1 71 17 0 0 0 63 27 0 0 0 69 18 0

1 0 66 12 0 0 1 62 42 0 1 0 64 1 0

0 0 64 17 0 1 1 74 4 0 0 0 72 25 1

1 1 70 1 0 1 1 66 19 0 1 1 59 29 1

0 0 64 30 0 0 1 70 28 0 0 1 69 1 0

1 0 78 1 0 1 1 83 1 0 1 0 69 42 1

1 1 75 30 0 1 1 77 29 0 1 0 79 20 0

0 1 70 12 0 0 0 69 12 0 1 0 65 14 1

1 1 70 1 0 1 1 67 23 0 0 1 76 25 0

1 1 78 12 0 1 1 77 1 0 1 0 69 24 0

1 1 66 4 0 1 0 65 29 0 1 1 60 26 1

0 1 78 15 0 1 1 75 21 0 0 0 67 11 0

1 0 72 27 0 1 0 70 13 0 0 1 75 6 0

1 0 65 7 0 1 0 68 27 0 1 1 68 11 0

1 1 67 17 0 1 1 70 22 0 0 1 65 15 0

1 0 67 1 0 0 1 67 10 0 1 0 72 11 0

0 0 74 1 0 1 1 80 21 0 0 0 69 3 1

;

proc logistic data=Neuralgia;

class treatment(param = ref ref='0')sex(param = ref ref='0');

model pain (event = '1') = Treatment Sex Age Duration/ rsq lackfit;

run;

proc logistic data=Neuralgia;

class treatment(param = ref ref='0')sex(param = ref ref='0');

model pain (event = '0') = Treatment Sex Age Duration/ rsq lackfit;

run;

eilayyan · Posted 02-19-2012 04:16 PM

Thanks

Owis