A scale reliability of .39 (and even .53!) is very low. Even if your main focus is not on the psychometric properties of your measures, you should still care about those properties. Inadequate reliability and validity can jeopardize your substantive results.

My recommendation would be to examine why you get such a low alpha value. Most importantly, you should first check whether each scale (item set) can be seen as unidimensional (measuring a single factor). This is usually done by running a confirmatory factor analysis (CFA) or item response theory analysis. Unidimensionality is a prerequisite for a meaningful interpretation of Cronbach's alpha (alpha is a composite reliability index for essentially tau-equivalent measures). CFA allows you to test the assumption of unidimensionality/essential tau equivalence and to examine the item loadings.

Also, you can take a look at the item intercorrelations. If some items have low correlations with others, this may indicate that they do not measure the same factor (and/or that they contain a lot of measurement error). Another reason for a low alpha value can be an insufficient number of items.

Hi,

You can run a normality check on each of the items on PD scale and assess the distribution. Probably one such item departs significantly from normality. That item probably was worded in a circuitous way. However, we have to use the scale as such, unless you are doing a validating study.

I agree with Christian Geiser that you need to do more to understand why you have such a low alpha. One common reason is negative correlations within the item set due to "reversed" items -- which violates the assumption in alpha that all correlations are positive.

I have found that a low Cronbach's alpha is often caused by the questionnaire design . I agree with Christian Geiser . If the number of questions to measure the item is two small a low CA can be produced . The use of the scale reliability CA if item deleted ( option) , can only be used until two items are left . Similarly even if using Omega to measure scale reliability the situation is worse since it calls for more than four individual questions to ensure the reliability of the scale . Prior validation of the scale through the theoretical underpinnings of the construct to be measured would give you an indicator as to whether it is unidimensional or multidimensional . I have learned do not accept concepts that are thought to be unidimensional when they are actually multidimensional .