Negative combach's alfa in evidence based questionnaire

#1
Hey everyone,

For a small research I had to do I used the Rosenberg Self-esteem scale and tried to look for correlations with aggression using the Buss Perry Aggression Scale.

Even though it proved very much insignificant I also found a negative crombach's alfa linked to the rosenberg scale.
I checked the coding which was correct. Does anyone have any idea what I'm doing wrong?

http://www.freeuploadsite.com/uploads/13694236301.zip

You can find the syntax and dataset in this file.

Cheers,

Gent.a
 
#3
Small research:

N=30


Item-Total Statistics
Scale Mean if Item Deleted Scale Variance if Item Deleted Corrected Item-Total Correlation Cronbach's Alpha if Item Deleted
RSS1R 27,4000 7,214 ,426 -,709a
RSS2R 29,0000 11,931 -,352 ,026
RSS4R 27,4333 8,599 ,079 -,368a
RSS6R 28,6000 10,524 -,190 -,085a
RSS7R 27,4667 8,189 ,187 -,470a
RSS3R2 28,8667 12,257 -,394 ,052
RSS5R2 27,1667 8,075 ,194 -,485a
RSS8R2 27,8000 8,993 -,030 -,255a
RSS9R2 27,2667 6,685 ,410 -,796a
RSS10R2 28,6000 14,110 -,563 ,222
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.


Inter-Item Correlation Matrix
RSS1R RSS2R RSS4R RSS6R RSS7R RSS3R2 RSS5R2 RSS8R2 RSS9R2 RSS10R2
RSS1R 1,000 -,276 ,219 -,213 ,389 -,135 ,476 ,176 ,420 -,068
RSS2R -,276 1,000 ,078 -,053 -,283 ,147 -,445 -,641 -,131 ,552
RSS4R ,219 ,078 1,000 -,196 ,128 -,312 ,024 -,002 ,278 ,003
RSS6R -,213 -,053 -,196 1,000 ,043 ,229 -,121 ,160 -,155 -,254
RSS7R ,389 -,283 ,128 ,043 1,000 -,582 ,524 ,398 ,518 -,712
RSS3R2 -,135 ,147 -,312 ,229 -,582 1,000 -,226 -,398 -,139 ,201
RSS5R2 ,476 -,445 ,024 -,121 ,524 -,226 1,000 ,380 ,420 -,543
RSS8R2 ,176 -,641 -,002 ,160 ,398 -,398 ,380 1,000 ,286 -,456
RSS9R2 ,420 -,131 ,278 -,155 ,518 -,139 ,420 ,286 1,000 -,504
RSS10R2 -,068 ,552 ,003 -,254 -,712 ,201 -,543 -,456 -,504 1,000
 
#4
Reliability Statistics
Cronbach's Alphaa Cronbach's Alpha Based on Standardized Itemsa N of Items
-,241 -,209 10
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.
 

spunky

Can't make spagetti
#5
would you be so kind as to clean up the SPSS data file that you uploaded and ONLY leave the variables that you're using to calculate alpha?

i'd like to try a couple of things on your data.

(ps- it's so awesome to have graduated. now i have all the time in the world to do whatever i want, heh....)
 
#6
Heu,
Yesterday the library closed so this is a bit late but I'll clean up the data after work and post it here.
Thank you for considering to help ;)
 

Lazar

Phineas Packard
#7
Half the Rosenberg scale is negatively worded. You need to reverse code those items. Not that it is relatively common that this scale factors out to two scales one consisting of the positively worded items and one the negatively worded items. Some people call the two factors self-esteem and depression.
 

spunky

Can't make spagetti
#8
Half the Rosenberg scale is negatively worded.
i didn't know that, but it makes a lot of sense. when i was looking at the datat the OP posted there seemd to be a chunk of negatively-correlated items on one part of the correlation matrix and a chunk of positively-correlated items on the other.

so i'm guessing it's a re-group and re-code kinda deal now...
 
#9
In SEM they always stress the dangers of some questions being coded negatively and others positively. I never understood why that mattered until now.

Will this effect factor analysis and or regression as well? We never checked in our surveys if questions were positively or negatively worded.