Attribute GLM (Generalized Linear Models) with Minitab

[milind dhakad]

Dear Barbara,
Good afternoon,
I have one question related with GLM, I have the data having attribute data .But to use GLM i have converted the attribute to numerical like okay job 100 and rejected job 0 and i have done the GLM
Pls find attached herewith the data and GLM analaysis
General Linear Model: C5 versus C6, C4, C3, C2, C1

Factor Type Levels Values
C6 fixed 2 in, ot
C4 fixed 5 1380-1360, 1390-1380, 1400-1390, 1410-1400, 1420-1430
C3 fixed 2 o, r
C2 fixed 2 o, r
C1 fixed 3 a, b, c


Analysis of Variance for C5, using Adjusted SS for Tests

Source DF Seq SS Adj SS Adj MS F P
C6 1 333 512 512 0.28 0.602
C4 4 18000 9692 2423 1.34 0.298
C6*C4 4 7333 15160 3790 2.10 0.129
C3 1 12000 13407 13407 7.42 0.015
C2 1 1667 2410 2410 1.33 0.265
C1 2 1419 1419 710 0.39 0.682
Error 16 28914 28914 1807
Total 29 69667


S = 42.5102 R-Sq = 58.50% R-Sq(adj) = 24.78%


Unusual Observations for C5

Obs C5 Fit SE Fit Residual St Resid
3 100.000 100.000 42.510 0.000 * X
13 100.000 100.000 42.510 0.000 * X
25 100.000 31.586 25.475 68.414 2.01 R
30 0.000 -0.000 42.510 0.000 * X

R denotes an observation with a large standardized residual.
X denotes an observation whose X value gives it large leverage.

I am not using Mintab so pls give the interpretation for above and does i had done the right thing by changing the attribute data to numeric value??
Or request you to guild me so that i can analyis the Attribute data output with variable and attribute inputs
Thanks
Milind

 

[Marc]

Another quick "Bump". My Thanks in advance to anyone who can help with this one.

 

[Miner]

I have one question related with GLM, I have the data having attribute data .But to use GLM i have converted the attribute to numerical like okay job 100 and rejected job 0 and i have done the GLM
Pls find attached herewith the data and GLM analaysis
General Linear Model: C5 versus C6, C4, C3, C2, C1

Factor Type Levels Values
C6 fixed 2 in, ot
C4 fixed 5 1380-1360, 1390-1380, 1400-1390, 1410-1400, 1420-1430
C3 fixed 2 o, r
C2 fixed 2 o, r
C1 fixed 3 a, b, c
Analysis of Variance for C5, using Adjusted SS for Tests
Source DF Seq SS Adj SS Adj MS F P
C6 1 333 512 512 0.28 0.602
C4 4 18000 9692 2423 1.34 0.298
C6*C4 4 7333 15160 3790 2.10 0.129
C3 1 12000 13407 13407 7.42 0.015
C2 1 1667 2410 2410 1.33 0.265
Error 16 28914 28914 1807
Total 29 69667
S = 42.5102 R-Sq = 58.50% R-Sq(adj) = 24.78%

I am not using Mintab so pls give the interpretation for above and does i had done the right thing by changing the attribute data to numeric value?? Or request you to guild me so that i can analyis the Attribute data output with variable and attribute inputs

Your data was not attached. Please attach and we can tell more. It is very difficult to provide proper guidance without the data.

One thing I can say is that your model is over fitted. That is, you have too many terms in the model. You can tell this because R-Sq is 58.5% while R-sq (adj) is much lower at 24.78%. R-sq increases every time you add a term to the model, while R-sq (adj) is reduced when you add a term that does not contribute to the model.

Factor C3 is the only factor that appears to contribute to this model based on the p-value of 0.015.

 

[AdamP]

HI.

If you've taken an attribute output (can't tell without seeing your data) and have converted it to numerical (100 for OK and 0 for reject), you may be better off running a logistic regression. If you only have 2 or so levels of the output, a GLM will try to solve for an unattainable solution somewhere in the middle. The logistic regression will tell you the probability of getting an "OK" result from the factors you include. Even with that, the previous responses are correct in having too many factors in the model.

Cheers,

Adam

 

[milind dhakad]

Dear sir,
Thanks for your reply and it is really helpful for me .
I dont have data available with me today ,But i will sent it to you as early as possible
Regards
Milind