See http://comparingcronbachalphas.org

Hi Akinsipo,

I noticed that you seems to work in SPSS environment. First thing which comes to my mind is do you really need to compare the alpha reliabilities of these three groups. Maybe you do. You may be interested in reporting that the scale/score can discriminate the cases within each of the three groups sufficiently. On the other hand, you may be interested in estimating the reliability in the WHOLE sample (and to note that the test instrument can discriminate the cases well) and then just compare the groups without thinking whether the cases in the individual groups can be separated from each other or not. That would be quite normal procedure when comparing, for example males and females - in many cases we do not see it necessary to estimate the reliability for those groups separately.

Second, If you ARE willing to report the reliabilities separately, in the SPSS environment, you use DATA > SPLIT FILE command. Then you select either COMPARE GROUPS or ORGANIZE OUTPUT BY. There may be some statistical procedure for really comparing statistically the values but I'm not that much aware of those. Usually, I guess, it is enough to use some heuristic measurement stick (low - sufficient - high) to assess whether the values differ from each other. Statistical tools could be found from estimating the confidence intervals for the alphas. You could use Feldt's procedure (1965) or Barchard & Hakstian (1997) and Hakstian & Whalen (1976) procedures for the confidence intervals.

Thank you Eduardo for the link. helpful. However am using SPSS and the values of cronbach for the three scales is not obvious. I have 52 items, divided into 133 subscales and finally making three scales (DP, SAA, SA).

Thank you Jari for the reply. you mentioned and I quote "On the other hand, you may be interested in estimating the reliability in the WHOLE sample (and to note that the test instrument can discriminate the cases well)" This is what I would really like to do if it is acceptable since am using three different groups.

Hi Akinshipo. If I understand you clearly, your response variable observed is "cronbach values" from treatment having three levels or groups. You are to use ANOVA statistical method to make the comparison. Since you are interested in comparison the three groups to identify probably the best, you will fit fixed anova model. You can do that using lm function in R. If you are a SAS user, you may use PROC GLM to fit the linear model. Having imported your data into SAS, invoke GLM procedure as

proc glm data=dataname

class group

model cronbach=group/ss3;

lsmeans group/adjust=tukey;

run;

quit;

if your design is CRD, use the above script in SAS but if RCBD, you have to incorporate the design or block effect into the model otherwise your error variance will maximize.

Thank You Moshood. Am using SPSS. I want to compare the cronbach of clinical, preclinical and graduates. is it really important to compare their cronbach like jari mentioned?

Akinshipo,

If your INTEREST is to compare the alphas, then Moshood's recommendation is relevant. If you interest is to compare the MEANS (of some interesting dependent variable) of the groups then there is not much idea/need of testing whether the cases can be separated within the different groups. However, testing the differences of alphas may enrich your own knowledge of the dataset. Think about the hypothetical situation that in your experiment one group gives ultimately (and consistently) low values, one group (consistently) mediocre values and one groups (consistently) high values. You would get the result that all the alphas in each group are close zero but in the whole sample it would be high. The zero alphas are of no value to report because that is not important but actually very awaited possibility caused by the design. What matters is that the test score in the whole data can discriminate the cases optimally in three levels.