Often, a patient’s condition is measured through combining individual scores on an ordinal scale. There is a whole host of possibilities as to what the scale may refer to here, including level of psychological distress, personality type or class of psychiatric disorder. For example, in assessing level of psychological distress, separate scores could be assigned to the patient according to the dimensions tension, anger, depression, fatigue and vigour. Based on the variation of the data within a given sample, you might then wish to assess whether the composite score for psychological distress (PD) based on these five measures is a reliable indicator of the level of PD by comparison with a hypothetical perfect PD scale applied to the population from which your sample was taken or by comparison with all hypothetical alternative PD scales applied to your sample. Reliability in these senses can be estimated using a statistic known as Cronbach’s alpha. Furthermore, based on sample data, one can assess whether the value of Cronbach’s alpha improves when any one of the individual components of the composite scale are removed.
Examples of the implementation of Cronbach’s alpha in medical research can be found on downloading a pdf version of the document at: Statistics notes: Cronbach’s alpha.
In turn, should you wish to calculate values of Cronbach’s alpha and specific inter-item correlations, to see which items appear to be most closely correlated, for your own data, you can find instructions on how to do so using SPSS at:
As the above resource explains, “[t]he type of reliability [intended here] is called internal consistency reliability: the degree to which multiple measures of the same thing agree with one another”. The same resource also provides a slightly amusing example to illustrate the individual steps required in SPSS. To enable you to get some practise, the relevant SPSS spreadsheet is also provided.
You should also consider obtaining a confidence interval alpha. The following video explains how this can be done in SPSS:
Further help with interpreting Cronbach’s alpha
By considering how Cronbach’s alpha is defined, you can get a better handle on how to interpret your findings. For example, take a look at the Wikipedia Resource on Defining and interpreting Cronbach’s alpha. You will see that the last formula in the section Definition tells you that Cronbach’s alpha may be viewed as the variance in the true scores for the individuals divided by the variance in the observed scores (which also includes measurement error). The larger the amount of measurement error, the smaller the value of Cronbach’s alpha and thus the lesser extent to which your questionnaire items are capturing the dimension they are supposedly designed to capture. When reflecting on this, note that:
a) Cronbach’s alpha increases with number of items, so maybe a low value for this statistic points to the possibility of extending the number of items in future work, not just that the items have been chosen imprecisely;
b) even so, by convention, the item total correlation statistic and alpha-if-item-deleted statistic are examined to see if it is possible to design an optimal set of items for any given dimension based on those items currently at your disposal (see the 2nd of the two pages in the resource Reliability analysis in SPSS to see how you might approach this rather simply);
c) you might like to consult the section Internal consistency of the above Wikipaedia resource to obtain a key for interpreting your values for Cronbach’s alpha on a scale from unacceptable to excellent
and
d) for the purpose of your reference list, the latter key is reported as having originated from George, D., & Mallery, P. (2003). SPSS for Windows step by step: A simple guide and reference. 11.0 update (4th ed.). Boston: Allyn & Bacon.
Please note that reliability (or, reproducibility) of scores is a necessary but not a sufficient condition for validity of scores (that is whether the scores provide a reasonable representation of what they are intended to). Here, scores are values forthcoming from an instrument such as an assessment or questionnaire and may therefore take the form of examination marks or disease severity, by way of illustration. Validity can take many forms. You may have heard of a few of these, including face validity, criterion validity and construct validity. Even construct validity itself falls into a range of sub-categories of validity, although each of these relies on different methodology. While validity comes in different types, the underlying concepts are relatively straightforward. To verify which of these concepts apply to your study and to help you use them more confidently within the context of writing up your report, you are recommended to consider the book Health Measurement Scales: A Practical Guide to their Development and Use by Streiner and Norman.
This book is currently in its fifth edition.
If you have the fourth edition, please note that the various sub-categories of construct validity are highlighted on pp. 250 – 251 of Chapter 10, while further details on each are provided in subsequent sections of the same chapter.
If you are registered with the University of Edinburgh, you can consult the electronic version of this edition of the book and gain details of the loan status of hard copies via the University’s library discovery system, DiscoverEd.
Here are some reference details:
Health measurement scales : a practical guide to their development and use
- Title: Health measurement scales : a practical guide to their development and use
- Author: David L. Streiner
- Geoffrey R Norman
- Publisher: Oxford ; New York : Oxford University Press
- Publication Date: 2008
- Edition: Fourth edition
Cronbach’s Alpha: Assessing Reliability by Margaret MacDougall is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.
