Cronbach's alpha will not tell you whether your scale is uni-dimensional (i.e., whether all your items measure the same construct). For that, you should construct an Exploratory Factor Analysis. For that, I personally would recommend using the options to do a Maximum Likelihood extraction method with oblique rotation (i.e., correlated factors).

Because there have been so many questions here about both Cronbach's alpha and factor analysis, I have compiled some resources on those topics:

https://www.researchgate.net/post/How_can_one_help_people_create_simple_scales_from_Likert-scored_items.

cronbach's alpha reading of O.5 is low. The acceptable level is O.6 if i am not mistaken refer  Hair. But I would like to point that if you are doing qualitative study,  it is not necessary for you to support the validity with cronbach, there are other means of validity techniques you can use to support valid data.

Hi Najmul

Cronbach's  alpha is used to measure the reliability of the questionnaire.  Simply  it means that if you use the same questionnaire once again with a  sample  the can we get the similar results.  This measure we used in quantitative methodological approach.

In qualitative approach we do not use this Cronbach's  alpha as we do not have suitable data to calculate. All you have text and transcripts of the interviews.

Hope this would help.

Can I recomend the article "Making sense of Cronbach’s alpha" by  Mohsen

Tavakol, Reg Dennick, International Journal of Medical Education 2011.

Related questions can be found on researchgate, e.g., https://www.researchgate.net/post/What_should_be_the_Cronbach_Alpha_thumb_rule

In short, there is no threshold for a good Cronbach's alpha, the statistic varies with things.

Usually, cronbach alpha is compulsory for the quantitative analysis to determine how reliable your items. Maybe you can run composite reliability (CR) for your research

On Cronbach's alpha being compulsory for anything, I'd recommend reading the Standards for Educational and Psychological Testing, chapter 2, where is discusses reliability. It does not say this.

Cronbach'a alpha is the measure of the reliability and consistency of the sampling instrument  and examine whether all the data were measuring the same underlying construct. If your alpha results is 0.5, it is acceptable but weak and particularly in a qualitative study, you need to support the results with verbatim quotes of the respondents, 

you have some excellent answers to your question.  If I were to reflect on your challenges, 

1.  what software are you using to measure Cronbach's alpha?

2.  how many items (questions) are used to score the reliability or Cronbach's alpha

3.  if  the data has 5 questions that you have tested  would the deletion of one question raise your Cronbach's alpha?  Please note that this deletion of a question needs to be evaluated against the theoretic expectations (face validity) of the underlying construct.

4.  if you are using SPSS, and you have several questions to capture a construct, then click on the button "statistics" and after that click the button "Scale if Item is deleted".  This will tell you if the deleted scale item increases the overall Cronbach's.  However, take care that the deleted question may be important to the theoretic expectations for the underlying construct.  If  the number of items are less than 3 for measuring Cronbachs, the reliability score is going to be low.  

Good luck with you research.

The simplest "diagnostic" is to look at your correlation matrix among your items. Since Alpha is based on the average correlation, you need to determine if you have a low value because all your correlations are low, or whether you have a mix of higher and lower correlations that are getting averaged into a mediocre results. If you are in the later case, then you should omit the items that are dragging down the average correlation.

If you are using SPSS Reliabilities, it will give you some of this same information under the heading "Alpha if item deleted" -- where you would remove any items that would produce a higher alpha when that variable is deleted.

Hi,

Cronbach's alpha is commonly used as an estimate of the reliability of a psychometric test for a sample of examinees. It is most fall within the range of 0.75 to 0.83 with at least one claiming a Cronbach's alpha above 0.90.

 Cronbach's alpha                                 Internal consistency    

α ≥ 0.9                                             Excellent (High-Stakes testing)

0.7 ≤ α < 0.9                                    Good (Low-Stakes testing)

0.6 ≤ α < 0.7                                               Acceptable      

0.5 ≤ α < 0.6                                                  Poor

α < 0.5                                                    Unacceptable

In your question you have mentioned that the Cronbach's alpha is calculated as 0.5.  So, it is an unacceptable one. 

Please Senthilvel can I know the source of your information as these contradicted with what I use to understand on Cronbach alpha range for decision. Again, if 0.5 is poor, why should it be unacceptable.

I think the value depends on sample variation.

Cronbach alpha is applied to determine how reliable your items in questionnaire. If the cronbach alpha below than 0.50, it indicates that your items are inappropriate. According to Nunnally (1978), the cronbach alpha should be greater than 0.7.

Classical measurement theory suggests the .70 minimum value most often recommended.for an alpha indicates the measure is 70% true score and 30% "noise" or error.

However, look closely at the formula for Cronbach's alpha.It is highly dependent on item and measure VARIANCE.

That is, the formula is derived from the sum of item variance (the numerator) over the total measure variance (the denominator). If both of these values are low, it indicates respondents varied little in their answers to items or the overall test. This suggests your items, as written, are not sensitive indicators of score variance. For this reason, I would do a complete item review and consider whether they need revision or, whether more items need to be included to increase the potential for response variance.

Hi Hoda,

See a nice discussion in Research Gate., e.g.:

https://www.researchgate.net/post/Cronbachs_alpha_below_07

In SPSS reliability procedure add under statistics Descriptives for Scale if item removed. When alpha increases by removing an item, delete that item from scale and check if you manage to improve the situation. If that does not help look for other items with a high correlation with the items in the scale to be included.  Try an other measure like Cohen's kappa.

no, It is not possible.

There are situations where the computation of a Cronbach alpha is simply not appropriate.  For example, if you try to assess knowledge about a subject matter before and after a training or educational program, then you may obtain a Cronbach alpha of 0 at the beginning (indicating that people had no knowledge at all about this subject and were answering randomly), but end up with an undefined alpha at the end if your intervention is successful (everybody got correct answers). This is a situation where you are trying to apply a psychometric analysis on a edumetric test. Carver made this important distinction about 35 year ago.  

what other situations that cronbach alpha is not recommended to be used ?

@ Mohamed Abdel Khaliq Dwikat: if your indicator covers more than one dimension, if there is no correlation between items.

Is it possible to change the items position betewwn the domains or combining the domains to improve the internal consistency according to cronbach alpha? 

Cronbach's alpha will not tell you whether your scale is uni-dimensional (i.e., whether all your items measure the same construct). For that, you should construct an Exploratory Factor Analysis. For that, I personally would recommend using the options to do a Maximum Likelihood extraction method with oblique rotation (i.e., correlated factors).

Because there have been so many questions here about both Cronbach's alpha and factor analysis, I have compiled some resources on those topics:

https://www.researchgate.net/post/How_can_one_help_people_create_simple_scales_from_Likert-scored_items.

By definition, Alpha is a measure of internal consistency. So, if your data collection instrument (be it cognitive or affective in nature) covers many different aspects, you are bound to get a low coefficients. Perhaps, separate the items into a few sets according to their content (nature) and find alpha for each set. 

As already mentioned, Cronbach's alpha is a measure of internal consistency, that is, how closely related a set of items are as a group. A higher value means a relatively high internal consistency of the items measured. A value of .70 or higher is considered 'acceptable' in most social science research situations. Sometimes, increasing the number of items increases the Cronbach's alpha. Accordingly, increasing the number of respondents in your case might help.

http://www.ats.ucla.edu/stat/spss/faq/alpha.html

 Increasing the number of items does not guarantee higher alpha; it depends on the qualities of the additional items. In fact, when developing a scale, deleting poor items increases the coefficient.

Of course, there is no complete certainty that increasing the number of respondents will lead to higher alpha.

Please refer to Field (2009),  depending on the type of constructs you are measuring an alpha level of 0.5 can realistically be accepted. Especially if the study is in the area of psychology.

Field is an excellent reference for this question.

 The Cronbach's alpha be increase, since the replications number of measurement or rater be increased.

JL Fleiss: Design and analysis of clinical experiments, 2011 

Testing unidimensionality and Cronbach's alpha are two different operations. You can test the unidimensionality by an exploratory factor analysis. The rule is: your should only have one factor with an eigenvalue above 1. Otherwise you have to split up the set of variables into different subsets each of which should be unidimensional. If you have a unidimensional set of variables and alpha is to low you can try to add variables, it means increase the number of variables but without destroying unidimensionality.

Dear Hoda, I have had such a similar experience, and I needed to perform other tests as Bartlett esphericity, and Factorial Analysis. In my case, many constructs were overlapped in terms of their conceptual extensions. I do agree with Dr. Morgan's answer.

I would be interested to know more about the questionnaire that you are using. Has it been used in previous studies/languages/samples? What was the Cronbach in these cases? If it was acceptable, this could mean something about the context that you are investigating that decreases the coefficient. If it is a measure specific for this study, I would suggest reconsidering the grounds on which you created it. If the aim of the study is qualitative, one idea would be to use the answers provided to individual items as meaningful information, instead of trying to summarise answers to non-appropriate scales. 

All the answers are relevant and correct according to the subject matter or topic of the study. 

Researchers usually try to convince themselves to accept low values of Cronbach's Alpha, so that they can quickly get to the "real analysis part". And they try finding references to justify what they have come across (i.e., low Alpha value). And there are studies which undermine the importance of Cronbachs' Alpha (which can be cited). However, in my experience, it is better to improve the Alpha value, rather than finding an excuse to use the lower one. It takes less effort, as explained below, to improve Alpha than to redo an undesired finding that is highly likely if you continue with low Alpha value:

1. Add more items to your analysis which measured the same concept more objectively and more specifically, than the items you are currently using.

2. While preparing questionnaire, be as objective as possible. Subjective questions with many possible answers are likely to result in poor scale reliability (low alpha value).

3. If you do not have more items to add to the existing analysis, and you are unable to conduct a new survey, try finding Cronbach's Alpha with items deleted one by one). It can be done easily in SPSS (as I shall mention below). Deleting one of the items, sometimes, improves Alpha in case the survey respondents struggled to understand your questions on that item.

SPSS Procedure:

Analyze>Scale>Reliability Analysis>Statistics (check descriptives for item,scale, and scale if item deleted)>continue> make sure Model drop-down is set to Alpha, click ok.

You will find various Cronbach's Alpha values if an item is deleted from the analysis in ITEM-TOTAL STATISTICS.

I agree with Soh Kaycheng that the quality of the additional items matters alot. For additional information please read; Nunnally, J. C. (1978). Psychometric theory (2nd ed.). New York: McGraw-Hill.

Are your items on a multi-point rating scale or are they binary? If your response options are polytomous (multi-point), then Cronbach's alpha will underestimate the internal consistency of your assessment. You can improve the reliability by collapsing the number of response options (i.e., if your scale is 1-5, change it to 1-3).

sir my total population is 153. for the total population my sample size is 110 as per the table of yamane's. sir i am using the simple random sampling. sir please help me for the simple random sample. how can i justify to myself for this sampling technique. please sir help me i am confusing for this one..

Dr Najmul Hoda , please go through with Nunally (1970) which discussed about the accepted range of Cronbach Alpha.

Dall'Oglio et al (2010), claims that a 0.5 Cronbach alpha can be legitimate and acceptable with a short scale (i.e. few items). In Developmental evaluation at age 4: Validity of an Italian parental questionnaire, they write:

p421 "Generally, a Cronbach’s alpha value higher than 0.70 is considered

to be satisfactory; however, when there are fewer than 20 items, as is the case for each area of the QS4-G, a value of 0.50 is satisfactory".

I have 5 x 3 item scales comprising in total a 5 dimension (15 item) professionalism scale, based on Hall (1968). I have alphas above and slightly below 0.50. These 3 item scales are low reliabilities compared to the 5 -item dimensions in other research (with other occupations, mine is further education). It suggests the 3-item dimensions have lost something of the validity in the 5-item scales, but I was testing a reduced scale. First, as the above (Dall'Oglio et al) is a published paper, I assume the 0.5 is acceptable to the academic community, and second, can the alpha coefficient be any lower than this for a 3 item scale? does anyone have any research they have come across to indicate this? Your help much appreciated.

Have you looked at removing each item in turn from the scale to see if this results in a higher Cronbach Alpha? It might be that one item loaded from the EFA erroneously or was cross loading, and that a smaller item set may better represent the scale.

Thanks for your reply. The standard professionalism questionnaire used in other research is 25 items, made up of 5x5 item scales, returning a consistent set of high cronbachs. The one I used was suggested in two previous studies which I was testing. Each individual dimension has 3 items. Removing items to create two item scales lowers the reliability score further. So for me, it's getting an authoritative position on what the lowest acceptable coefficient is for a 3 -item scale. Dall'Olglio et al have stated 0.50 is satisfactory, but is there other evidence that suggests one could go lower?

my scale reliability was very low like it was .3 and .4. how can i justify my low reliability of the scale through useful citations which i can used as a reference in order to justify the low reliability of my scales. please suggest some authentic citations, books or articles which i can use as reference for the justification of my low reliability.

As far as my concern, when the cronbach alpha falls in the range of 0.3 to 0.7, we may take the data for further analysis. Because. 0.1 to 0.3 is one extreme of giving responses and more than 0.7 is an another extreme of responses. So avoid the data of which it is lower than 0.3 and greater than 0.7

Thank you, this is really helpful, I don't suppose you have a citation for this at all?

Kindly refer the article"diffusion on innovation in social networking sites among University Students" - O. Folorunso............ In International journal of computer science and security. Volume(4):Issue(3).

If the reliability shows 0.5. Do reversible score in Spss. Go to transform menu click compute variable, there you have to type =max score+1-item name to improve ur reliability. But it is unethical, and you need to know this is one of the way to improve the reliability value