When running a Cronbach's Alpha, you want to aim for a coefficient score of 0.70 or higher. However, it is difficult to get a good score when the scale you are using has a low number of items. Assuming that you are using SPSS to run your Cronbach's Alpha, you should go back in, go to analyze, then scale, then reliability assessment, then click on Statistics. Here you will be able to check the box to see which sub-scale you need to remove to get a score of 0.70 or higher. Once you find that sub-scale, simply remove it from your data set, and run your Cronbach's Alpha again. Good luck!

When running a Cronbach's Alpha, you want to aim for a coefficient score of 0.70 or higher. However, it is difficult to get a good score when the scale you are using has a low number of items. Assuming that you are using SPSS to run your Cronbach's Alpha, you should go back in, go to analyze, then scale, then reliability assessment, then click on Statistics. Here you will be able to check the box to see which sub-scale you need to remove to get a score of 0.70 or higher. Once you find that sub-scale, simply remove it from your data set, and run your Cronbach's Alpha again. Good luck!

Hi Elisa,

The ATHI consists of 4 sub-scales. I tried to do what you suggested for one of the sub-scales, 'Personal Causation;. The 'Personal Causation' sub-scale's alpha is .470 and is made of 3 items. On SPSS, I clicked on the "if deleted" button and deleting any of the three would not help much.

Deleting the first item would only decrease the alpha lower to .251

Deleting the second item would only decrease the alpha lower to .322

Deleting the last item would only increase the alpha lower to .528

The subscales sound too small to me. It also sounds as if the scale itself, with which I am not familiar, is rather small to split into subscales and is not homogeneous with this population. Even so there are many scales of say 12 or so where you can get scores of 0.7+ I am afraid I would conclude that you are not measuring a stable construct with this scale with this population. That in itself is quite interesting if other researchers have found something different. Two solutions, both of them difficult I'm afraid would be to try another scale or to get enough data for factor analysis. But I understand both of those options may be unrealistic or not open to you. You could also try to find the questions in the scale which have the lowest correlation with the others. That is also potentially interesting. Another option is to do some cognitive testing of the scale with the group if you have not already done so, to understand why they are responding to the questions in the way they are and this might lead to some insight into the underlying constructs. Sorry if this is all unhelpful - I don't know enough about the research or what you are trying to achieve.

Hi Allen. I am not surprised that the coefficient score of that sub-scale is low because it has only three items. Please ensure that you use all four sub-scales of the ATHI to run your reliability analysis. Also, you should check to ensure that there were no reverse-scored items that you overlooked and did not apply the correct scores to. If your items in your survey instrument are not properly scored, this will result in a low Chronbach's Alpha coefficient score as well.

Hi Andrew William McCulloch

Thank you for your very comprehensive response. I've been reading more and more and other researchers have also faced similar issues with reliability, albeit have had slightly more success. The original authors, Kingree and Daves (1997), reported the internal consistency to be .71 for the total scale, .73 for the societal causes subscale, .72 for the personal characteristics subscale, .65 for the affiliation subscale, and .60 for the solvable problem subscale. . The fact that they also found a .60 score suggests there have always been problems with the scale. In other people's research (e.g. Ruggerio, 2015 and Chancellor, 2010) they've also had to remove items and have alpha scores of around .6-.7.

This has made me think that the scale isn't very good and might need adapting. I have now found out that Snow-Hill (2019) has developed the scale further, in light of the low reliability. Unfortunately, I did not know about this and can now only mention this in my discussion. The questionnaire was a research project as part of my degree to investigate attitudes towards homelessness across social-work and computer-science students. It was an anonymous survey and we also can't redo the ethics form. This means that I can't redo the scale (as the deadline for the project is May) and I am not sure what you mean by collecting data for factor analysis - my supervisor has informed me that data collection has finished. Unfortunately we can't conduct any cognitive testing like you suggest, though I will mention this in the discussion!

Thank you anyway!

Hi Elisa Douglas

When we included all four sub-scales of the ATHI, this still had an alpha score of .582. As far as I am aware, I have reversed the items and included the reversed items in the reliability analysis :/ I've attempted doing this fresh three different times and my course-mates have also come to the same conclusion. :/

Yes thanks. My comment about factor analysis is not very helpful in this context. It is the way of understand the structure of the scale when you have a very large data set. It would normally be done on all scales to be used regularly in say clinical or educational settings. It sounds as if you are right about this scale. I would say that if you present the analysis you have summarised here that should put you in good standing for this dissertation, after all this type of research is about finding things out and demonstrating that, not necessarily choosing the "perfect" scale, which does not exist anyway - and relies on the underlying construct actually existing of course! Which is a whole different issue, and what do we mean by our constructs anyway................. Good luck.