For example for a 17-item scale (where the Xs mark deleted items) you end up with

Items Pre Post Post-then

K1

K2 X X

K3 X X X

K4

K5 X X X

K6 X X X

K7

K8

K9 X X

K10

K11 X X X

K12 X X X

K13 X X

K14 X X X

K15 X X

K16 X

K17

Alpha 0.91 0.89 0.888

1 factor- 8-items 7-items 9-items

Hello, if you're trying to test reliablity, I advise you to run Principal components analysis becasuse you'll get a better overview of item contribution to each subscale. You can choose from several rotation options. Probably varimax will be the best solution for you: it forces item saturation on a single factor with no losses of explained variance.

Then delete itens with loadings bellow .30

Run Cronbach's alfa again

Thanks Edgar - I have run the PCA - just surprised that I am getting three very different sets of items each time - same respondents - so wanted to make sure I was right in doing it this way.

Hello,

That's very normal. Be aware of negative loadings, since they may indicate you'll need to invert item-scale. That can be a serious problem since it lows your cronbach's alfa

Hello Edgar,

Thanks for the advice, I reverse coded all the appropriate items - and dealt with outliers and missing data. Any other ideas?

If you are steel struggling with low cronbach alfa, and:

- Eliminated low loading items

- Dealt with missing cases

- Reversed items

There’s not much to do, unless to admit that your data does not go along your theory...

Yes uses the Cronbach's alpha (if item deleted) & factor analysis to reduce scales on three separate sets of data - or going with the items kept after the first analysis. But if your number of observations ins less than 30 use cluster of cases (not variables, because you don't have at minimum 50 observations (Mallotra). The cluster of cases is a procedure similar to factorial analysis when the number of observations is