Dear all,   I am posting a question regarding reliability analysis in behalf of my colleague.  His instrument has four subscales and each subscale has four items. The respondents are asked to use 0 to 100 to rate each item within each subscale such that the sum of the four items is 100.  Can we use the Cronbach alpha for the reliability at each subscale level?    For your further info we tried to compute the cronbach alpha of the first subscale with n=42 using spss.  The alpha is negative and out of range.  Please see spss output below:   Thank you. Eins.   Reliability Statistics  

Cronbach's Alpha(a)

N of Items

-203.120

4 a  The value is negative due to a negative average covariance among items. This violates reliability model assumptions. You may want to check item codings.                                           Item Statistics  

Mean

Std. Deviation

N

ME_A

36.31

13.300

42

ME_B

20.76

8.453

42

ME_C

20.43

9.698

42

ME_C_A

22.86

10.250

42                                                           Item-Total Statistics  

Scale Mean if Item Deleted

Scale Variance if Item Deleted

Corrected Item-Total Correlation

Cronbach's Alpha if Item Deleted

ME_A

64.05

177.120

-.992

-.791(a)

ME_B

79.60

72.491

-.980

-6.281(a)

ME_C

79.93

88.751

-.984

-4.473(a)

ME_C_A

77.50

114.939

-.988

-2.969(a) a  The value is negative due to a negative average covariance among items. This violates reliability model assumptions. You may want to check item codings.                                           Scale Statistics  

Mean

Variance

Std. Deviation

N of Items

100.36

2.918

1.708

4