Logistic regression

[asanquer@virbac.fr]

Hello,

I need some help with the logistic regression:

I have data on the test of 8 devices.
The result of the test is a factor with 3 classes (bad, correct, good).

I would like to use a logistic regression
to analyse the potential difference between devices on the test result.
So I transform the test result in a
binary variable (bad vs correct or good), and the device variable in 8
binaries variables (device1, device2 , ... device8).

I perform my logistic regression using
7 device variables to test each device versus the 7 others. 

summary(glm(TEST ~ DEVICE2 + DEVICE3
+ DEVICE4 + DEVICE5 + DEVICE6 + DEVICE7 + DEVICE8 , DATA, family=binomial))

to test the Device1 versus the others,

summary(glm(TEST ~ DEVICE1 + DEVICE3
+ DEVICE4 + DEVICE5 + DEVICE6 + DEVICE7 + DEVICE8 , DATA, family=binomial))

to test the Device2 versus the others,.....

Firstly, I would like to know if it
is a good way to test the differences between devices.

And secondly, I have a problem when
some devices presents no bad test results:

Coefficients:
         
        Value  Std. Error     t value

(Intercept) -22.4782974 501.9650441
-0.04478060
    DEVICE1  -0.6132228
  0.3126136 -1.96159988
    DEVICE2  -0.8047190
  0.3040136 -2.64698379
    DEVICE3  -0.4645971
  0.3071465 -1.51262350
    DEVICE5  -8.7565248
317.4695750 -0.02758225
    DEVICE6  -8.7565248
388.8192084 -0.02252081
    DEVICE7  -1.4746061
  0.5543023 -2.66029210
    DEVICE8  -1.9182577
  0.5447448 -3.52138796

Here, the devices 5 and 6 had 0 bad
results although the device4 which is tested had 57.7% of bad results.
But they have huge std error, which induce a non significant difference
between them and the device 4.

This results is not logical. So, is
there a special way to analyse this type of data ?

Thanks for your help

Annaële Sanquer