Cronbach’s Alpha is the most commonly used statistic for determining the internal consistency of measurements, such as items in a questionnaire, exam, or survey. It is equivalent to performing the split-half methodology on all combinations of these items.
Hi Charles
Your article is informative and good for beginner.
I have index developed with 14 components with score zero or one, for which I have obtained data using responses of more than 100 questions.
So my question is, can I use cronbachs alfa for 14 components or all 100 questions while applying the test
Hi Monica,
You can certainly use Cronbach’s alpha for 100 questions. Regarding 14 components, this depends on what these components are.
Charles
Like everything else on this site, the section on Cronbach’s alpha is awesome!
I am working on an interactive/refreshable item analysis report using PowerBI (another Microsoft software that has a lot in common with Excel). To do this, I have been writing DAX codes to create dynamic calculations of many of the coefficients, indices, and values you discuss throughout the sections on reliability and item analysis. The dynamic aspect is that they recalculate the values depending on how you filter the data, basically like pivot tables and charts in Excel.
For instance, my report can filter for different versions and different sections used in our battery of test tasks.
The issue I have is that in some of the filtered contexts, I am getting negative values for Chronbach’s alpha. To be sure I had not made an error in my code, I have extracted the data for one such filtered context and used your analysis tools. I was happy to see that I got the same value!
This negative value cannot be due to a coding issue. The answers are correct or incorrect (0, 1) and all of the data has been exported from the software we use to administer computer-based exams. So it is acurately coded and there are no data entry errors.
This must mean that there is an issue with discrimination, correct?
In fact, I have two items (out of ten, for this particular task) with discrimination values of -0.49 and 0.20, repsectively. When I filter these items out, I get a positive alpha value (albeit a poor one), which is confirmed by your analysis tools.
So poor discrimination must be the issue, right?
Hi Mike,
Thank you very much for your kind words about the Real Statistics website.
It is entirely possible that Cronbach’s alpha takes a negative value. This clearly indicates poor internal consistency.
It seems logical that this can be caused by poor discrimination, but I have not investigated whether this can be caused by other factors.
Charles
Hi Sir,
I want to measure the non-financial performance of the firm with customer satisfaction, employees satisfaction, employees loyalty, and product & Service quality with five point Likert Scale. The questions are;
1. Can I use the five point likert Scale?
2. Can I collect data for the same items from firm owners/managers?
Hi Inam,
1. Yes, you can use a 5-point Likert scale. This is a common practice.
2. Yes, you can collect data from anyone you choose, including owners and managers. When you do any analysis you need to be careful since the results from people in the same firm may be correlated.
Charles
Hello Charles
I have figured out the missing data covariance analysis using excel functions only
With the missing data can you still do the factor analysis as discussed in the basic concepts page
Thanks again
I am not sure how to calculate the reliability using Cronbach Alpha – I have 13 respondents and
I have the following answers – Agree
Disagree
Agree
Agree
Strongly Agree
Agree
Agree
Agree
Strongly Agree
Neither Agree or Disagree
Strongly Agree
Agree
Neither Agree or Disagree
Please help me to find if the study is reliable or not? and the score please
Thank you in advance
Hi Jeanette,
You need to assign a Likert scale to your possible responses. For your data you could use:
Strongly Disagree = 1
Disagree = 2
Neither agree or disagree = 3
Agree = 4
Strongly Agree = 5
Now you can calculate Cronbach’s alpha as described on the webpage.
Charles
Hello,
I administered the questionnaire and now I am waiting for a response. I need to find out which questions are the most important because I want to reduce the number of questions in the questionnaire? How can I do that?
Ekica,
How do you define “most important” in this context?
Charles
Hi Charles
I have conducted a pilot study for my research work ” Emotional labour on healthcare providers”. I used Cronbach’s Alpha for the 5 point Likert scale where I am getting a negative result. Can I use -3,-2,1,2,3 as the scores for strongly disagree, disagree, neutral, agree and strongly disagree instead of 1,2,3,4,5.
Santa,
Yes, if you use -2, -1, 0, 1, 2, but no if you use -3, -2, 1, 2, 3 since 0 and 1 are skipped.
-2, -1, 0, 1, 2 is equivalent to 1, 2, 3, 4, 5.
Charles
Hi Charles,
I am developing a questionnaire on Learning styles measuring 3 factors, Visual, Auditory and Kinesthetic through self reported ranking.
There are 5 questions under each of the factor with 3 choices of ranking, which means a total of 15 questions and 45 choices.
Example Question
I spend free time by…
a) Watching television or videos
b) Talking to friends
c) Doing physical or sporting activities
For each of the 15 statements, participants have to rank their choices as 1, 2 or 3 which gives me 3 ranking for each of the 15 questions.
My queries are…
1) Is Cronbach Alpha the right measure of internal consistency?
2) If yes, what is the process?
Hope I have made my question clear? Thanks in advance
Jose,
Your question is clear, but as I stated in my previous response, I don’t know of a good approach. It is likely that a separate measurement would be required for each of the 3 factors (as would be the case when using Cronbach’s alpha), but I don’t know what the appropriate measurement would be.
Charles
Hi Charles,
I am carrying out a survey regarding financial literacy and I have broken down the questions into three subsections. The first subsection contains two questions in which the participant has to calculate and type out the answer with a specific value being the correct answer, one multiple choice question with only one correct answer while the rest are incorrect, and one true/false question with only one correct answer. Each correct answer holds a value of 1. Would it be possible to test the reliability of this subsection with Cronbach’s alpha?
Amir,
As long as each of the questions is measuring the same concept, you should be able to carry out Cronbach’s alpha. Even though there are different types of questions, the value for each question is 0 for a wrong answer and 1 for a correct answer.
Charles
Hello, charles,
What formula would be best in calculating the reliability of a matching type items of 20 thanks.
James,
What do you mean by “the reliability of a matching type”?
Charles
Hello Charles,
I’m a bit confused! should I calculate alpha for the piloting sample only or for the whole sample responses? and What if my piloting sample is only 15, is alpha the suitable test? if not what do you suggest?
Thank you in advance.
You might use Cronbach’s alpha on a pilot sample prior to giving a survey or questionnaire to a larger sample. In this way, you can detect any reliability issues that require you to modify the survey or questionnaire prior to rolling it out to a larger sample. Cronbach’s alpha can be used, however, on the larger sample.
Cronbach’s alpha can be used on a small sample, but the larger the sample the more certainty you will have about the results; more specifically the smaller the confidence interval for the Cronbach’s alpha value. See
https://www.real-statistics.com/reliability/internal-consistency-reliability/cronbachs-alpha/cronbachs-alpha-continued/
Charles
Thank you very much.
Hi Charles,
With regards to calculating inter-rater reliability, could you please explain why Cronbach’s alpha is not suitable when you only have 2 raters, but ICC’s are acceptable for this?
I cannot seem to find an explanation for this online anywhere.
Best wishes,
Gerard
Gerard,
You can perform Cronbach’s with only two “raters” (for example two questions).
Charles
Hi Charles,
Thank you for your response. I am not sure my question was very clear – I am talking about assessing the internal consistency of the ratings of two judges (who rated participants’ creativity on a 1-7 scale).
I have read from a few sources that Cronbach’s alpha is not suitable for this when there is only 2 judges, but that ICC’s can be used for 2 raters.
Does this sound correct to you?
Again, thanks very much for your help.
Gerard,
Cronbach’s alpha is used for internal consistency and not for agreement between raters (which ICC is used for). In any case, Cronbach’s alpha can be used with two items and ICC can be used with two raters.
Charles
Thank you for your help, Charles.
Gerard
Hello!
I am here to ask if how could I asses the reliability of a questionnaire if number of items, frequency and percentage is only the data I have? Is it possible to solve for its cronbach’s alpha value? Below are the links of questionnaires I wish to test for reliability.
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5871321/table/table2-2515690X17746547/?report=objectonly
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5871321/table/table4-2515690X17746547/
I hope you could help me this concern. Thank you!
Joy,
Yes, in fact, you only need the frequency data to calculate Cronbach’s alpha (percentage and counts can easily be derived from the frequency data).
Charles
Thank you very much for your rely!
Hi
I am trying to access the reliability of online programming tests. I have candidate scores who have taken the same test only once. The test can have multiple questions. I also have individual scores of each candidate for every question. How can I use Cronbach’s alpha in this scenario? Is there any other better method of doing this?
For example one of the tests which access the candidate coding skills in java has the following information:
Number of questions:12 (9 MCQ’s and 3 coding questions)
Number of candidates who took the test: 44
Candidate scores:[]
Candidates individual score for each question:
Max marks for MCQ question=5
Max marks for coding question 1(hard level)= 30
Max marks for coding question 2 and 3(medium level)= 20
Please help me how I can access the reliability of such a test. Thanks
Sach,
For each MCQ does a person get 5 points if the response is correct and 0 points if incorrect or is there partial credit?
Combining coding questions and MCQs introduces a complication. Reliability requires that the same “concept” is measured for all questions. This assumption may be violated. If so you can create a separate Cronbach’s alpha for the MCQs and the coding questions. You also need to decide whether you need to use the same weights for the coding questions (i.e. does the 30 point question count more than the 20 point question)? Also are these coding questions measuring the same concept.
Charles
Hey Charles,
Thankyou so much for replying!
So for the MCQ’s, there is no partial credit. It’s either 0 or 5. And yes a 30 point question counts more than a 20 point question. Yes, these coding questions measure the same concept.
So I have calculated the Cronbach’s separately for both MCQs and Programming questions. And I am getting really low value for most of the tests. What is a solution to resolve the low values. Is there anything else I can try– the parameters, the formula, etc.
Sach,
If they are measuring the same concept then you could probably combine the questions assuming no partial credit. Use the coding 1 = correct and 0 = incorrect.
The low Cronbach’s alpha can be due to a variety of factors, including poor consistency (reword the questions), incorrect coding, the questions really don’t measure the same concepts, etc. This issue is explored further on this webpage.
Charles
Sach,
This is the usual situation for Cronbach’s alpha. There are other approaches, but Cronbach’s alpha is commonly used in this situation.
Charles
What does Cronbach’s alpha implies?
just wanna hear it from you because I’m having trouble understanding my professor in online classes 🙂
See the second paragraph of
https://www.verywellmind.com/what-is-reliability-2795786#:~:text=Reliability%20refers%20to%20the%20consistency,should%20be%20approximately%20the%20same.
Charles
Hi Charles,
Can i use cronbach even my instrument is not likert scale? My questionnaire is answerable by yes or no plus there is a duration (time and motion). Thank you
Cristy,
Yes, you can.
Charles
I am using a tool with 5 dimensions and each dimension has different number of items using the same scale. Should I calculate Cronbach Alpha for each dimension? Can I find Cronbach Alpha for the whole tool takin all the items of all 5 dimensions together or the average of the Cronbach Alpha of the 5 dimensions?
Is there a method to find the validity of the tool.
Nancy,
Cronbach’s alpha should only be used to determine the reliability of a questionnaire that is evaluating one concept. This seems to indicate that you will need to calculate a separate Cronbach’s alpha for each dimension. You can take the average of these measurements or use the minimum of them or something else. This is up to you, but it is not clear how you would use such a combined value.
Charles
Hello Charles,
For my independent variables in my questionnaire, some questions use the 5 point likert scale of disagree/agree and some use 5 point likert scale of unlikely/likely, will I have to do one Cronbach’s alpha for the disagree/agree questions and one for the unlikely/likely questions, OR do I just combine them and do a cronbach’s alpha for each independent variable, regardless of the scale as they are both 5 point?
I hope this makes sense, thank you!
Hello,
If the two types of questions are measuring different concepts then you will need to do two separate Cronbach alphas. If the two types of questions are measuring the same concepts and disagree/agree and unlikely/likely have similar meanings in relationship to this concept, then you only need to do one Cronbach’s alpha.
Charles
Thank you for replying! So if my concept was brand recognition and i had questions with likert scales of disagree/agree and unlikely/likely, then I would only do one Cronbach’s alpha for this concept regardless of the likert scale, is that correct?
Hello Garsin,
It depends on whether “brand recognition” is indeed a single concept.
Charles
Hello,
is it possible to compute the Cronbach’s alpha if the choices are ‘Yes’ and ‘No’? Because when I compute it, the value will become negative
Thank you.
Pong,
Yes, but you need to code Yes as 1 and No as 0.
Charles
Hi,
I would like to understand in my QUESTIONNAIRE, there are few questions which are NOT APPLICABLE for certain respondents to answer. Hence it is UNFILLED.
So while running reliability test (CRONBACH ALPHA), how should I code those NOT APPLICABLE questions.
Request to kindly answer
See
https://real-statistics.com/reliability/internal-consistency-reliability/cronbachs-alpha/cronbachs-alpha-with-missing-data/
The following might also be interesting:
https://www.researchgate.net/publication/328567140_Calculating_the_Cronbach's_alpha_coefficients_for_measurement_scales_with_not_applicable_option
https://stats.stackexchange.com/questions/77125/cronbachs-alpha-with-missing-data
Charles
Hi Charles,
I am trying to measure the internal consistency of a self made questionnaire. I have three parameters. On each parameters I have different set of questions with different scales. But the scales in themselves are very different. For example one of the parameter is financial knowledge of participants. So in one question i am asking them to calculate the best credit score with options like 850,900,715 and 652 . While on the other question I am asking them if they have missed an EMI payment before with options like yes, no, once.
Is it even appropriate to use cronbach’s alpha to this kind of questionnaire?
Hello Neha,
The key issue is that all the questions must be measuring the same concept. Otherwise, you need to calculate separate Cronbach alpha values for each group of questions where each group represents a single concept. Assuming this is done, then if different scales are used then you need to make sure that the scales are consistent: e.g. if question 1 uses values 1,2,3 and question 2 uses values 1, 2, 3, 4, 5, then assuming that 1 from Q1 is equivalent to 1 from Q2, 2 from Q1 is equivalent to 3 from Q2 and 3 from Q1 is equivalent to 5 from Q2, then you need to redefine the 1,2,3 scale for Q1 to 1,3,5.
Charles
Hello Charles,
I am wondering if I can use Cronbach Alpha to ensure reliability and validity for A/B Split testing. I am testing 2 email versions against each other (different subject line for opening rate and different button for click-through rate).
I appreciate your time and consideration.
Hello Franziska,
Cronbach’s alpha is used for measuring internal consistency reliability (not validity). If you have a questionnaire to help you with A/B testing then Cronbach’s alpha can be useful. The basic goals of A/B testing will be achieved by other statistical tools.
Charles
Hello Sir! How can I apply cronbach alpha on my risk assessment survey? The risk are categorized (design risk, equipment risk, etc.) and under each category there are corresponding risks which the respondents will rate the likelihood (scale 1-5) and impact (scale of 1-5) of the risk. I’m using an objective scale. Thank you!
Cris,
You probably need to calculate two Cronbach alpha scores: one for likelihood and the other for impact.
Charles
Thank you Sir Charles!
I have another question, if I calculate each Cronbach Alpha scores for likelihood and impact, should I consider it per risk or per risk category?
What if my Cronbach Alpha score for likelihood is below 0.70 which unacceptable, and my Cronbach Alpha score for impact is above 0.70 which is acceptable, what does it mean to the risk score ( RS=Likelihood*Impact)? Does it mean that the risk score is not reliable?
Thank you sir!
Hello Cris,
1. Sorry, but I don’t understand your first question.
2..70 is commonly viewed as the cutoff, but this isn’t universally the case.
3. It is entirely possible that one portion of a questionnaire or other instrument is reliable and another part is not reliable.
Charles
Hi Charles,
I have 3 different models. And I used a between subject design to test them such that 36 people worked with these models independently, i.e., 12 people worked with each model. I asked them 3 questions to measure their trust in the model they worked with in a 5-point Likert-type scale.
I measured alpha but it isn’t so high. I cannot increase the items and deleting items doesn’t make big difference. Do you have any suggestion how can I increase the alpha? Does have low number of participants any affect on alpha?
Thanks!
First of all, you need to make sure that you are calculating a correct value for Cronbach’s alpha. In particular, you need to make sure that you aren’t measuring multiple concepts and that you employ reverse coding if necessary.
If this is taken care of, then you need to look carefully at the questionnaire (or whatever instrument) that you are using.
Theoretically, the number of participants shouldn’t affect the value for alpha, although it will likely impact the variability of alpha.
Charles
Thanks Charles,
I have another question too. In my case which test can be used to see whether three populations have the same distribution?
I was thinking of t test but I think it is not possible since the population size in each model is 12 and it is not normal distribution.
Then I though about Wilcoxon signed-rank test, but since this is a between subject I don’t think it is possible (I’m not sure)!
Can you please suggest me a valid test to check the population distribution?
have a nice day,
Elahe
To determine whether three samples come from a population with the same mean, you would use ANOVA. If the data is not normally distributed, you might use the Kruskal-Wallis test or Welch’s ANOVA.
You could use Anderson-Darling’s two-sample test to determine whether samples A and B come from the same population. You could then use this test again to determine whether samples B and C come from the same population (and so all three come from the same population); you should use an error correction factor since you would use two tests.
All of these tests are described on the Real Statistics website and are supported in the Real Statistics software.
Charles
HI Charles,
I have developed my own tool, which has no likert scale, it asks respondents to choose 7 words from a list of 85 words, to best describe themselves. These words came from a qualitative research i had done some years ago. How can i calculate the reliability and validity here. Cronbach gives me a negative value.
These words have not value, they are only positive and negative words, which respondents dont know. but it is at the back end.
I must say i am not a SPSS or expert researcher. I have been doing all this on excel will help to get your views, as a lay person.
Thank you
PG
Hi PG,
Cronbach’s alpha only measures reliability and not validity.
A negative value for Cronbach’s alpha is possible and indicates very poor reliability, but it is important that you check to see whether you have measured Cronbach’s alpha correctly before you reach this conclusion.
First of all, you need to code the response correctly. E.g. if you have two questions that measure the same thing except that one is phrased positively and the other negatively (called reverse coding), then you need to correct for this in the coding of the responses, otherwise the Cronbach’s alpha value won’t be correct. This issue is explained on the webpage along with other issues.
One potential approach is to code each of the 85 words with a 1 if that word is selected and a 0 if it is not selected. But if the word is negative, perhaps you need to use a code of -1 instead.
Charles
Thank you Charles. That was helpful
Hi Charles,
In my calculation, I have a negative Cronbach’s alpha value.
No. of items/questions= 20
Sum of the item variances= 19.89375
Variance of Total scores= 12.2375
Cronbach’s alpha=- 0.65857
what should I do now?
It is possible to get a negative value, but unlikely to get a value as negative as -.65857.
A negative value indicates a very poor outcome, usually requiring changes to the questionnaire, however, there are other reasons that might cause such a low value as described on the webpage (e.g. reverse coding, a questionnaire that is testing multiple concepts, etc.)
Charles
Hello sir, good day. I just want to ask how much sample (number of respondents) should I get if i have 30 items 5point likert scale questions (strongly agree, agree, neutral, disagree, strongly disagree) to test the validity of my self made questionnaire? Thank you.
See the following webpage re the size of the sample to test reliability
https://real-statistics.com/reliability/internal-consistency-reliability/cronbachs-alpha/cronbachs-alpha-power/
Charles
Hi Charles Greting from India
My research study title is “Develop and validate quality of life tool for the children diagnosed with epilepsy”
I had collected response from 194 patient in the form of Likert scale in various section like Physical, Social, Cognitive, Psychological and Treatment. Where Always meaning is 1 and Never meaning is 5. As example
You are supervised by parents and family members. (Always(1), Frequent(2), often(3), Very rare(4), Always(5). But I am getting negative Cronbach Alpha for all 5 domain. How can I convert it in positive. Help me.
Hi Vipin. Glad to communicate with someone from India.
The usual reason for getting a negative result is an error in the calculations or an error in applying Cronbach’s alpha.
If you have conducted the calculations correctly, then a negative result would indicate low reliability.
But before drawing this conclusions make sure that you don’t have any reverse coding and the questions are all measuring the same concept. These issues are discussed on the website.
Charles
Can cronbach alpha can be used to calculate the consistency of an NBA player performance?
I guess it is possible depending on how you measure the performance of the player.
Charles
Hi Charles,
I have done Cronbach’s alpha for 2 types of Likert scale in my study. I would like to know if I have to do Cronbach’s alpha test if the answers are like:
– Yes
– No
– Unsure
Also if I am to re-code the values can I recode yes = 1, No = 2 and unsure = 3
This approach would imply that unsure is ordered higher than no, which may not be appropriate. The following paper suggests a different approach.
https://www.researchgate.net/publication/328567140_Calculating_the_Cronbach's_alpha_coefficients_for_measurement_scales_with_not_applicable_option
I plan to add this to Real Statistics shortly.
Charles
Thank you Charles
Hi,
I need help urgently.I am measuring the stress levels and their effect on performance by keeping emotional intelligence and resilience in mediation.The statements related to stress are positively worded i.e I am able to complete my work on time,I am able to devote time to my family.The likert scale coding for these statements related to stress are 5 is for Strongly agree to 1 for strongly disagree.The statements for emotional intelligence are like “I am able to appreciate other’s emotions”.The coding is 5 is for Strongly agree to 1 for strongly disagree.Resilience has statements like “I am able bounce back from stressful conditions within a short period of time”.The coding is 5 is for Strongly agree to 1 for strongly disagree.If I take cronbach alpha for each dimension seperatley,I am getting alpha greater than 0.7.If I combine all the dimensions,it is coming negative.For stress,if the value is high that means employees have no stress.For EI,if the value is high it means employees are highly emotional intelligent and can handle stress easily.There is a negative relationship between EI and stress.My query is how to correct alpha if all th dimensions are considered?
Regards,
S.Srivathsani
1. In general, you need to change the coding so that all the questions are either positively or negative worded. When using a 5 point Likert scale, you reverse coding by using 6-s instead of s as the score.
2. You should create separate Cronbach’s measurements for questions that measure different concepts. In your case, perhaps one for emotional intelligence and another for resilience in mediation.
Charles
Hi Charles,
The questions are positively worded for stress.I have changed the the likert scale to 5 for strongly disagree and 1 for strongly agree.Emotional intelligence is also positively worded.The scale is 5 for strongly agree to 1 for strongly disagree.The problem is when I put all alpha together i am getting negative covariances error.
Thanks for all the help.
As I have said, you shouldn’t put the two together as they are measuring different concepts.
Charles
Hello, charles
I have calculated the cronsbach’s alpha using your first example but I get the value of less than 0.5
If you email me an Excel file with your results, I will try to see why you are getting a different answer.
Charles
hi. i am using the Vocabulary Knowledge Scale for my thesis and i want to run the reliability test for it. the VKS is actually a self report assessement, however each answer will be given a score. for example, when the word given is “beautiful”, they need to rate the word according to the scale :
1- i never saw or heard the word.
2- i saw or heard the word but don’t know the meaning.
3- i saw n heard and i think the meaning is_____.
4- i know this word and the meaning is _____.
5- i can use it in a sentence. _______(make a correct sentence)
so my question is, if the student answer’s is 4 but he gave a wrong answer, should i enter the data in spss according to what he answered?
i hope my question is understandable.
Hello Titi,
You first need to decide how you want to score the answers. For example, you can (1) score the answers using a Likert scale of 1 to 4. Alternatively, you can (2) score a correct answer as 1 and an incorrect answer as 0. You might score the answer as 0 if incorrect and 1 to 5 using the Likert scale if correct (this is simply a Likert scale of 0 to 5).
Once you decide this, you can use Cronbach’s alpha (at least with any of the above codings).
Charles
Hi Charles, I want to measure cronbach’s Alpha to check the reliability of my questionnaire. There are 27 questions (divided into three parts Conclusions, Inferences and Identifying Bias) and the number of participant are 444. but the problem is to measure the Conclusion skills there are 2 passages (1st passage consist of 5 questions and 2nd passage is containing 4 questions) followed by 9 questions with 3 options. 2nd part Inference is consisting of 3 passages followed by 9 questions (3 questions in each part) with 3 option. In part three (Identifying Bias), there are 9 questions but only one questions has 3 options and other 8 questions has 4 options. Could you please suggest me the way to organize my data and check cronbach’s Alpha as a whole.
Best regards
Uddin
Hello Uddin,
I don’t completely understand the situation that you are describing, but here are some observations which may help:
1. If the Conclusions, Inferences and Identifying Bias parts of the questionnaire are measuring three different concepts, then you should calculate three separate Cronbach alpha values based on the questions that test the appropriate concept.
2. If you have some questions which use a 5 point Likert scale (1,2,3,4,5) and others a 3 point Likert scale (1,2,3), I would change the coding for the second type of question to align with the first type, e.g. by changing 1,2,3 to 1,3,5. Dealing with a 4-point scale perhaps the coding 1,2,3,4 maps to 1,2.33,3.67,5. I have not checked to see how important these mappings are, but this approach makes sense.
3. If instead of a Likert scale, you have multiple choice questions with a correct answer and one or more incorrect answers, then use the coding 1 for a correct answer and 0 for an incorrect answer.
Charles
Dear Charles,
I am carrying an experiment very similar to your Example 1, in which I have several subjects that answer to a 20 questions questionnaire on a Likert scale.
I used Cronbach’s Alpha to evaluate internal consistency.
After a period of time, all subjects are submitted to the same questionnaire and again I compute Cronbach’s Alpha.
Is there a way to evaluate the repeatibility of the questionnaire as a whole?
I mean, I could compute the ICC(2,1) for each of the 20 questions to evaluate the repeatibility of each question, but I wonder if it exists a statistical method that considers all the questions together and gives a “global repeatibility index”
Thank you very much for your help
Piero
Piero,
You can look at this as a test/retest case. Simply calculate the correlation between x1,…,x20 and y1…y20 (e.g. using Excel’s CORREL function) where xi = the sum of the scores for question i by all students on the original test, while yi = the sum of the scores for question i by all students on the retest.
This does not involve Cronbach’s alpha. In fact, Crombach’s alpha is used to avoid having to do test/retest.
Charles
Dear Charles,
thank you for your answer. To compute correlation, I used the Real Statistics tool, I get a very high r=0.98.
Since scores for questions can have only values 0,1 or 2, what do you think is the best correlation coefficient to use (Pearson, Spearman, Kendall)?
I have also a methodological question. By using the sum of scores of each question, I can’t consider individual variations, that is, I could get the same sum of scores from many different combinations of individual answers and this would lead to r=1 even if single subject’s answers did change.
As I can understand, instead such individual variations are taken into account if I compute ICC(2,1) for each of 20 questions.
There isn’t an ICC generalization that takes into account all question scores?
Thank you again
Piero
Hello Piero,
1. It really depends on what you will do with the answer, but generally I would suggest using Kendall.
2. ICC(2,1) is calculated using test results for all 20 questions and retest results on the 20 questions.
Charles
Hello Charles,
about point 2, perhaps the problem is that I’m not understanding well how to arrange data in Excel.
Could I send you my spreadsheet, so you can have a look and help me to arrange data in the correct way?
Thank you very much for all your valuable help
Piero
Piero,
Yes, please send me an Excel file with your data and results.
Charles
Piero,
Yes, you can use ICC(2,1) as well since this is another way of conducting test/retest. I haven’t written this on the website yet, but I will do so shortly.
Charles
Hello, is it possible to compute the cronbach’s alpha if the choices per item is a, b, c, or d? Thank you.
Yes,
Code the correct answer as 1 and any incorrect answer as 0.
Charles
Thank you so much
Btw, can i compute the cronbach’s alpha using pilot testing or i still need to have a re test? Thanks.
You can estimate it during the pilot. In fact, that is the usual approach. You don-t need a retest.
Charles
I’ve done with the computation and i got a value of 0.69756, how can i interpret this?
There isn’t uniform agreement as to how to interpret Cronbach’s alpha. Usually .7 is viewed as acceptable.
Charles
thank you so much. this article/site help me a lot.
Based on some references, the alpha of 0.69756 is low (questionable). how can i interpret this value if my items are 50, respondents or variables are 30, and my scale is just 1 or 2. thanks
Thank you so much Sir Charles! 😊😊
Objective of cronbach’s alpha method
Internal consistency reliability
Charles
hi charles,i am very new in python.i want to measure the consistency f data set. caan u plz sent me the code for testing internal consitency??
I don’t don’t use python, but the Excel explanation on the webpage gives you a pretty idea about how to code the python version.
Charles
I have set of users(user1,user2) answering the set of questions(question1,question2) for the set of subjects(subject1,subject2). Please let me know how should I proceed to calculate the Cronbach alpha for the whole set.
Kunal,
Cronbach alpha should be used only for questions that ask about a single concept. You should calculate a separate Cronbach alpha value for those questions related to each concept (probably subject in your context).
Charles
Dear Dr. Zaiontz,
I need to use Cronbach’s Alpha to check/prove/calculate the reliability of my test. I have gone through Test/Re-test method/procedure and I have administered the test twice with an interval. According to KR21, the reliability is 0.917 and 0.919 for test and re-test respectively. I have used an online tool to calculate this.
Would you please explain to me how I can apply Cronbach’s Alpha to this? And would you please tell me if I have done the right thing to calculate once for the “test” and once for the “re-test” regarding the KR21?
Thank you so much and lots.
My test has 50 questions and 56 students have participated. Thanks so much and lots.
Arrange your data in a range with 56 rows and 50 columns and apply the Cronbach’s alpha test as described on the webpage.
Charles
hello Charles,
I am using cronbach alpha for testing reliability of my tool, the tool consist of 8 sections ad every section has different number of items. the issue is in section one the likert scale measuring is like ; poor to excellent. while for some sections it is like : strongly disagree to strongly agree. so i have to test cronbach alpha for all at once or individually for each section.
Hello Saeeda,
If the questions are not measuring the same concept then you should not try to calculate one Cronbach’s alpha value for all the questions. Instead you should calculate separate Cronbach alpha values for each group of questions, where one group of questions corresponds to one concept.
If the problem relates to the scale, then you can probably deal with this by making the scales the same. E.g. if you are using a scale of 1,2,3 for one question and 1,2,3,4,5 for another question, you could change the scale of the first question to 1,3,5 (or whatever you find suitable).
Charles
Is there any literature support available for measuring alpha for different factors separately?If we know that the questions are measuring different constructs,we need to measure alpha separately.I need literature support for this.Please help me with this.
Almost any article about Cronbach’s alpha will state this.
E.g. this is addressed in the following Wikipedia article
https://en.wikipedia.org/wiki/Cronbach%27s_alpha
Charles
Hi
Sorry for teh trouble.Can you provide me some authentic source as wikipedia is editable?I need to put this reference in my thesis
You can use the references in the Wikipedia article. Also, after a few minutes of Google search, I am sure that you will find a suitable reference.
Charles