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.
Topics
- Basic Concepts
- Real Statistics Support
- Hypothesis Testing and Confidence Intervals
- Power and Sample Size
- Handling Missing Data and Not Applicable Entries
- Proof of Main Property
References
Tavakol, M., Dennick, R. (2011) Making sense of Cronbach’s alpha. International Journal of Medical Education
https://www.ijme.net/archive/2/cronbachs-alpha.pdf
Gliem, J. A., Gliem, R. R. (2004) Calculating, interpreting, and reporting Cronbach’s alpha reliability coefficient for Likert-type scales
https://scholarworks.indianapolis.iu.edu/server/api/core/bitstreams/976cec6a-914f-4e49-84b2-f658d5b26ff9/content
Newsom, J. (2017) Empirical estimates of reliability: Cronbach’s alpha
https://web.pdx.edu/~newsomj/pmclass/empirical%20estimates%20of%20reliability.pdf
Trizano-Hermosilla, I., Alvarado, J. M. (2016) Best alternatives to Cronbach’s alpha reliability in realistic conditions: congeneric and asymmetrical measurements
https://www.frontiersin.org/journals/psychology/articles/10.3389/fpsyg.2016.00769/full
When coding variables on a likert scale it is common to use the numbers 1, 2, 3, 4, 5 and 6 (etc). But in some cases the numbers 0, 1, 2, 3, 4 and 5 (etc) are used. This can affect the Alpha found in some cases. Example:
How much are you working how much time per day (%) do you wear glasses?
a) 1 to 20%,
b) 21 to 40%,
c) 41 to 60%,
d) 61 to 80%,
e) 81 to 100%,
f) I do not know or I prefer not to respond.
The option “f) I do not know / I prefer not to respond”, this could be either code zero or code 6.
In this case is there any recommendation for encoding responses?
Is there contraindication to using Cronbach’s Alpha?
Should I delete who replied “f) I do not know / I’d rather not answer” to do Cronbach’s Alpha?
Hi Sofia,
I don’t see how you can compare choice f with the other choices (unless of course you are quite sure what choice f represents). I would simply leave these responses out of the calculations.
Charles
Thank you Charles!
Hi Charles,
I am testing both validity and reliability (Cronbach alpha) of a questionnaire that has a different number of response options per items (I have a combination of -5, 6, 7 response options in my questionnaire).
I am wondering whether I can transform all the scores into standard z-scores (in SPSS) and then perform principal component analysis and calculate Cronbach Alpha?
Thanks in advance.
Hello Zara,
If you are using Likert scales with 5, 6 and 7 scales, then I can’t see any reason why you could make then all the same. E.g. you could make then all in the interval 0 to 1. E.g. 1,2,3,4,5 becomes 0,.25,.50,.75, 1.0. Similarly 1,2,3,4,5,6 becomes 0,.2,.4,.6,.8,1.0.
Charles
In the SPSS you have this option:
>Analyze
>Scale
> Statistics
> Inter-item > Correlations
or
> Sumaries > Correlations
Them you will have the “Cronbach’s Alpha Based on Standardized Items” on your output.
Hello Charles,
Thanks a lot for your guide, much appreciated. I was wondering if you could provide some additional explanation of how you obtained the results in Figure 3 (leaving out Q1 in column B, Q2 in C, etc.).
From a conceptual standpoint, I am having difficulties understanding how can respondents’ answers change when one question is left out.
Thanks in advance for the help!
Andrea
Hello Andrea,
The respondents answers don’t change when you leave one question out. You simply recalculating Cronbach’s alpha eliminating the one question from the calculations. This gives you some idea how much that one question is contributing to the Cronbach’s alpha score.
Charles
Hi Charles,
I used Cronbach’s Alpha in my Post-Study System Usability Questionnaire (PSSUQ). PSSUQ, a 19-item questionnaire has 3 sub-scale(System Usefulness(items 1-8), Information Quality(items 9-15), Interface Quality(items 16-18)) and Overall scale(items 1-19). (Reference: https://www.researchgate.net/publication/220302199_Psychometric_Evaluation_of_the_PSSUQ_Using_Data_from_Five_Years_of_Usability_Studies)
My result was:
System Usefulness: 0.8392
Information Quality: 0.7696
Interface Quality: 0.7459
Overall: 0.9109
My question is; why is the Overall higher than the three sub-scales? Should it be between 0.74 and 0.83?
Hello Anthony,
If you are checking the reliability of 3 different concepts, then there is no sense in calculating an overall reliability score since it won’t have any meaning. I don’t know why it comes out higher than the other values.
Charles
Hi Charles, thanks for such a comprehensive explanation of Cronbach’s Alpha, but there is one more thing I would like to understand. What are the theoretical foundations or practical justifications for using population variance and not the sample variance as the denominator of the last element of the presented equation (figure 2, cell B21)?
Best regards,
Oskar
Hi Oskar,
You can use the sample variance (in all places) instead of the population variance and you will get the same answer.
Charles
Hi Charles,
I conducted a Study in which we measure people’s personality ratings by looking at a picture of their face. I’d like to test the consistency of people’s responses (their ratings of the picture) by trait (Openness, Conscientiousness, Extraversion, Agreeableness and Neuroticism), do you think doing C’s alpha for this would be enough?? If not, do you have an alternative recommendation?
Thanks for your time!
– Bel
Hi Bel,
It depends on what you mean by “consistency”. Are you looking for a measure of the agreement among multiple raters on these picture based on the 5 categories that you have listed? What sort of rating are you using? a Likert scale?
Charles
Hi there,
I have conducted a survey, however some questions are a 4 point likert scale and some are dichotomous. Can i test the reliability of both? I have tried but the numbers dont seem right, I wondered whether they all have to be the same scale?
Hello Grace,
You need to make the scales similar. Perhaps if the 4 point Likert scale is 1, 2, 3, 4, then a dichotomous rating could be 1 and 4, provided the 1’s have a similar role and the 4’s have a similar role.
In any case, the important thing is that the 4 point Likert questions and the dichotomous questions are measuring the same thing. If not, you need to calculate separate Cronbach’s alphas, one for the 4 point Likert questions and another one for the dichotomous questions.
Charles
Good day. I have developed an instrument for the pretest , then data are expressed in score, like, respondent 1 scored 32/40, and so on an so fort, having 32 respondents, is it okay to use Cronbachs alpha for internal consistency, thank you.
Ruben,
Based on the information you provided, I can’t say for sure, but if each respondent has only one score, then Cronbach’s alpha can’t be calculated.
Charles
Hi Charles,
I want to test the internal consistency of a tool that comprises 9 questions that must be answered on a four-level Likert scale. However, we have results from an inter-rater reliability assessment for 2 raters and 25 subjects. My question is: considering that I have results for 25 different subjects, which one should I use to calculate the Cronbach’s alpha? Or should I use an average score for the 25 altogether?
Hi Hudson,
Internal consistency and inter-rater reliability are two different concepts, and results from the latter are not relevant for the former. You can use Cronbach’s alpha if say 25 people filled in the questionnaire using a Likert scale of say 1,2,3,4. Don’t use averages, use the scores for each question from all 25 people.
Charles
What is the best tool to test the realibility of a 40 multiple choice questions with four alternatives.? Thank you.
It is debatable what the best test is, but the most commonly used such tool is Cronbach’s alpha.
Charles
Hi Charles! Hope you can help me. In my study, I used a questionnaire that uses a Likert scale of 1-5, 1 is poor;2 is fair; 3 is good; 4 is very good; and 5 is excellent. I used 2 groups of respondents since my study is to evaluate the performance of each group and correlate. The Group A composed of 47 respondents rated the performance of 7 respondents in Group B, while the 7 rated the performance of 47. A 48 item questionnaire is used for both groups. Now, can I use a split half correlation or cronbach’s alpha to test the reliability of the questionnaire using the average score per question?
Renelyn,
You can use the Split-half or Cronbach’s alpha to measure the reliability of the questionnaire, but I don’t understand what you mean by “using the average score per question”. These measurements use the actual scores and not the average scores, unless the each score given is the average of the 7 Likert scores given by each respondent in Group A or the average of the 47 Likert scores given by each respondent in Group B.
Charles
Hi Charles. If my variable involves some sort of measurement like measuring blood glucose, is this Cronbach’s alpha applicable?
Thanks.
Janet,
Probably not, but if it is a questionnaire about blood glucose, then maybe yes. It depends on the details.
Charles
Hi Chales,
Thank you for the explanation. I would like to know why ANOVA is used for KR – 20 calculation.
You don’t need to use ANOVA to calculate KR20 (which is a special case of Cronbach’s alpha), but it turns out that the calculations used to calculate Cronbach’s alpha can also be done by ANOVA. There is probably an inherent reason for this, but I haven’t had time to figure this part out.
Charles
Hello everyone! I have a question regarding on how to use crohnbach alpha because I have no idea at all on how to apply it in my research and I am new to this. I use likert scale 1-5 where in 1= poor to 5=excellent, my respondents are 196 and I have 32 questions that use likert scale and I want to test for 30 respondents for validity and reliability test. How can I know that my questionnaires are reliable and validate? If I use crohnbach alpha for my study, should I use crohnbach alpha 0.5 or what? Please help me here.
Nini,
Cronbach’s alpha can be used to calculate reliability. There isn’t universal agreement about what is a sufficiently high value for Cronbach’s alpha, but .5 would probably be considered to be too low. Usually .7 or more would be considered acceptable; some might accept a lower value.
This webpage explains how to calculate Cronbach0s alpha.
Cronbach’s alpha is not used for validation. In fact, demonstrating validity can be difficult. One approach is to show that the results from your questionnaire are comparable to some test that has already been validated.
Charles
If KR-21 is for example 1.247, how do you traduce. If reliability is until 1.0, why this value is greather than 1. Is KR-21 reliability?
Teresa,
KR21 is just an approximation for KR20. It was used in the days before calculators and computers. In any case, I don’t believe that the value should be larger than 1. It does measure reliability.
If you send me an Excel file with your data and calculation, I will try to figure out why you are getting a value greater than one.
Charles
how to cite this for my research?
Angel,
See Citation.
Charles
hello need some help. i do not understand how to compute B20 or cell M17 on figure 1
Mildred,
=SUM(B17:L17)
Charles
Can you please recommend the appropriate statistical methods to use in testing the following hypothesis?
1. No relationship between the availability and the frequency of use of electronic resources.
2. Library Service quality will have no significant influence on the level of library users’ satisfaction.
3. There is no significant difference between the users’ expectations and perceptions of the library service quality. Paired T-test is suitable for testing two related variables like ‘User’s Expectations and Perceptions’
Additional information is required to give a definitive answer to your question, although it appears that a paired t test is appropriate for #3.
Charles
Thank you all for your contributions, I also have a sample size 768 to study but contemplating on the actual size of pilot test samples to be selected for me to be able to use a Cronbach’s alpha. I have mixed questions in Section 1-5 of the instrument, Section 1 – has 5 Items, sec 2 has 9 Items. Sec3 = 8 Items , Sec 4 = 20 Items and sec5 is divided into two subsec. 1 has 9 and subsec2 has 16 respectively. Among them, only sec 4 & 5 have 4 Likert scales.
Considering the pattern of questions is it possible for me to use Cronbach’s Alpha? if the answer is yes can I chose 50 for a pilot test from 786 sample size.
Hope someone helps.
If all the sections are testing the same concept, then theoretically you can calculate one Cronbach’s alpha measurement. From a practical point of view, this also depends on what sort of coding you have for sections 1, 2 and 3. Are they some other type of Likert scale or are they questions that have a correct or incorrect answer (such as TRUE/FALSE or multiple choice questions).
If the various sections are testing different concepts, then you should calculate separate Cronbach alphas, one for each concept based on the questions that correspond to each concept.
You can certainly do a pilot based on 50 of 786 samples.
Charles
Thank you Charles for your response.
Hi Charles Thanks for this tutorial.
However I can’t find in my excel the formula for Cronbach’s Alpha?
Is there something I should install or activate with my excel first?
Thanks
Hi Sendelyn,
This is not an Excel function. You need to download and install the Real Statistics Resource Pack to gain access to this function. The download is free. See Free Download
Charles
How to test the reliability of cronbach alpha value with given factor loading value?
For example, given factor loading values are :
F1: .780
F2: .827
F3: .729
Cronbach’s Alpha value = ?
Are these factor loadings from Factor Analysis? If so, then you should use factor analysis to find the hidden factors and then map each of the original questions onto one factor. You should then calculate separate Cronbach alphas for each factor (i.e. on the questions that correspond to that factor).
E.g. if you have 3 factors and questions 1,2,6,8,9 correspond to factor A, then you should calculate Cronbach’s alpha based only on those questions. Similarly if questions 3,4,7 correspond to factor B then calculate a second Cronbach’s alpha using only those questions, and finally if questions 5,10,11,12 correspond to factor C, then calculate a third Cronbach’s alpha using only those questions.
Charles
Hi Charles,
Great tutorial for calculating Cronbach’s alpha. This is super helpful!
I have a question.
I’m doing reliability and item analysis of a test for Chemistry course.
There are three sections each with different types of questions in the test: 1) 10 Multiple choice question 2) 10 True or False questions 3) Open end questions with multiple points. Some questions are 3 points worth, some 5 points worth and some even 10 points worth.
Should I calculate the C-alpha separately for each section or do it once for all? If I have to just calculate once for all of the sections combined, do I need to convert section 3 questions with different multiple point values to a common scale ie 0-1?
Also, is it acceptable to calculate an alpha for section a and section b combined and an alpha for section c?
Thanks,
Sahil
Sahil,
True/False and Multiple Choice questions can be combined as long as all the questions are measuring the same concept. Both types of questions are rated 0 for incorrect and 1 for corrected. Open-ended questions use a different scale, and so you will have more trouble combining them with the other questions.
I guess you could change the scale of the the True/False and Multiple Choice questions to 0 for incorrect and 10 for correct, and double the scale of the open questions worth 5 points. You then have to decide how to handle the open questions worth 3 points: perhaps: 0 for 0 point, 3 for 1 point, 7 for 2 points and 10 for 3 points. You can now calculate Cronbach’s alpha for all the questions together. Again, the critical issue is that all the questions should measure the same concept.
Charles
α =0.69014 What this mean when i calculate 150 respondent for expectations of 22 SERVQUAL questioner?is my data ok? reject the null?
also tell me some where it is written p value must be p ≤0.05 and some where p ≤0.5
what is this
Jami,
There is no clear agreement as to what the value of alpha = .69014, although in general I would think that it would be viewed as borderline acceptable or just outside acceptable.
I have never seen p ≤0.5.
Charles
Hi Charles, I have looked down the questions and answers here, but I can’t find an answer to an issue I have. Is it possible to use the same question from a survey under different subsections, which are tested for internal consistency separately?
For example: Two of my subsections are:
1. Knowledge
2. Resources
My question on the survey is “I know how to use the resources I need to complete my academic work” (5 point Likert: SA, A, N, D, SD)
Can I add this question to both 1 and 2 when I am calculating the Cronbach’s Alpha scores for each of these subsections, or can a question only be used once (regardless of the number of subsections used)?
Many thanks for your help with this.
Craig
Craig,
You can use the same question on two different surveys and calculate Cronbach’s alpha for each of the surveys separately.
Charles
Many thanks for the quick reply, Charles. Very much appreciated. May I just clarify that I have understood your answer. I am using a single survey of 38 questions. The 38 items on the survey have been completed by over 400 respondents. The questions have been allocated to one OR MORE of 10 subsections, which I have tested individually using Cronbach’s Alpha. If I have understood your answer correctly, this is an acceptable procedure.
Apologies for the follow-up, but I just want to double check this is OK before I finalise my survey. Thank you again, I was really struggling to find a definitive answer to this question.
Craig,
It seems like you have understood correctly.
For each subsection, the value of Cronbach’s alpha for that section incorporates the value of the repeated question.
Clearly, you should be able to use the question in multiple subsections since I could create a survey tomorrow using this same question, and this would have no impact on your survey. The only caution I would add is that if repeating the question in multiple subsections influences the answer given, then this could conceivably cause a problem. E.g. if the person gets tired of answering the same question and then gives an arbitrary response or after seeing the same question in subsection 2, based on the influence of the questions in this subsection, goes back to subsection 1 and changes the answer to the question in that subsection.
Charles
Thank you, Charles. The question is only asked once but the analysis for each of the subsections will use some of the questions more than once so this should not be a problem. Many thanks again for your help and advice. Craig
Good Day!
I have 2-Group of respondents (10 School Heads and 20 Teachers). I am going to give them the same question. Do I need to use cronbach alpha seperately to test the reliability ? Thank You and more power!
Erwin,
You can combine all 30 and calculate one value for Cronbach’s alpha, but I think it would usually be more meaningful to calculate two separate values.
Charles
Dear sir, what do I do when I have missing data? In my database I have now written NA for the items which are missing, but obviously I can’t calculate Cronbachs alpha then.
I will clarify my question a bit more:
I have a empathy questionaire in which I have given to 14 children, but not everyone did answer every question in the questionaire, and I can’t delete all the child’s response. Furthermore, I can’t delete one of the questions, because the missing answers aren’t all for one question. In the ideal world, I would just love, if I can make a Cronbachs Alpha reliability test after all.
Ida,
You have two choices>
1. Eliminate all the questionnaires that have one or more missing items. This may mean that you have almost no questionnaires left.
2. Using some imputation process. If you have only a few missing items, you could use a simple imputation technique< otherwise you should use multiple imputation. These techniques are described elsewhere on the website. Charles
Hi, Would appreciate some help please. I am currently piloting the questionnaire that I am using for my thesis. It is a Protection Motivation Questionnaire, which has 6 sub-constructs (Fear, Self- Efficacy, Response Efficacy, Response Cost, Perceived Sensitivity and Perceived Severity). It is looking at what affects women’s intentions to attend cervical screening. At the end of the questionnaire there is a question (Intention) looking at whether they intend to attend screening or not?
I need to look at the reliability of the scale. So I calculated the cronbach alpha for all 6 sections separately (mostly at =.8 and =.6) and the scale as a whole (=.76). Whenever I calculate the cronbachs alpha, should I be including the intention question? Or just enter the items for each sub construct?
Also my second question is that one of the sub-constructs (Response Efficacy), has low cronbachs alpha and is displaying this error message: The value is negative due to a negative average covariance among items. This violates reliability model assumptions. You may want to check item codings.
There are no reversed item scores, so I am unsure what this error could be caused by.
Liza
Liza,
Yes, you should calculate Cronbach’s alpha for each of the 6 sections separately. If there is only one question for intention, then you won’t be able to calculate Cronbach’s alpha for that question (nor would its value have any meaning).
The message you are getting isn’t coming from Real Statistics so I don’t know how to comment on it. As far as “reverse coding” is concerned, it is important that all the codes go in the same direction. How to correct this is explained on this webpage.
Charles
Liza,
Yes, you should calculate Cronbach’s alpha for each of the 6 sections separately. If there is only one question for intention, then you won’t be able to calculate Cronbach’s alpha for that question (nor would its value have any meaning).
The message you are getting isn’t coming from Real Statistics so I don’t know how to comment on it. As far as “reverse coding” is concerned, it is important that all the codes go in the same direction. How to correct this is explained on this webpage.
Charles
I’m not sure what else to use to work out the reliability. Basically we have two variables: MeanTime: which is the average time that a candidates takes to complete the task, and ErrorTime which is how much of his run time was spent in an error state. But I’m only considered in the first variable MeanTime. It is measure in seconds so I have 500 lines that looks like this
Person A – 110
Person B – 56
Person C – 89
I’ve computed the Z-scores for this but would like to know the reliability.
Hi Charles,
I have a set of data in which i need to workout alpha for just one variable and its z-score. The variable however is a time measured in seconds. This is for a test that measures 2hand co-ordination. Is there a way to do this?
Matt,
Why do you want to use Cronbach’s alpha?
Can you give me some idea of what your data looks like?
Charles
Hi Mr.Charles
I faced the negative Cronbach Alpha problem for a 5 Likert scale.
So according to your recommendation:
-Do I have to subtract 6 from the negatively phrased questions?
-Can you give me a scientific paper support this ?
Thanks
If you have some positively phrased questions and some negatively phrased questions, you need to make sure that the 5-point Likert scales agree (e.g. 5 has the same meaning for all the questions, and similarly for the ratings 1, 2, 3 and 4).
Almost any scientific paper on Cronbach’s alpha will support this or it will be implied). The ratings don’t have similar meaning, Cronbach’s alpha is completely meaningless.
Charles
Hi, I want to insert Multiple Response Questions (MRQ) in the exist data in SPSS to calculate its realibility using Cronbanch’s alpha. Is it possible to do that ? And if yes, how can I insert the MRQ into the exist data in SPSS ?
Sorry BB, but I don’t use SPSS. This site is about using Excel to perform statistical analysis.
Charles
Hello! Mr Charles, Im conducting research using likert five scale, Sample size 87 with 10 question. Pls how do calculate the cronbach Alpha?
Chris,
The referenced webpage explains how to calculate Cronbach’s alpha. Do you have a specific question?
Charles
hi,
how to calculate cronbach’s alpha n coefficient as there’s 150 respondents n 6 questions are likert scale n 5 questions are option one (like tick one) . No of items means no of respondents or what.
Susan,
This is explained on the referenced webpage. Do you have a specific question?
Charles
Hi Charles,
I have a questions, I have a survey with likert-scale (12 questions for one construct) with only two respondents, and I tried to do the Cronbach Alpha using the formula =cronalpha from realstats, but it gives an error (#VALUE!). Please advice how can I do the cronbach alpha with only two respondents.
Sophia,
I just tried using the formula =CRONALPHA(A1:J2) to calculate Cronbach’s alpha with only two respondents and 12 questions. It worked fine.
If you send me an Excel file with your data and use of the CRONALPHA formula, I will try to figure out what is going on.
Charles
I just figured out that the error is not because of only two respondents. can you please help me how to calculate the cronbach alpha for this data below then. I tried using formula, or using the data analysis tool, or the calculation, but nothing gives the cronbach alpha without error.
Respondent 1: 5 5 5 2 6 5 5 1 1 7 7 7
Respondent 2: 4 6 6 6 6 6 3 4 2 2 4 7
Thank you!
Sophia,
Unfortunately, Cronbach’s alpha is undefined for this example since the sum of the scores for respondent 1 = 56 = the sum of the scores for respondent 2. Thus, the variance for these sums is zero. This is the value var(x_0) in Definition 1 on the referenced webpage, and so the calculation of Cronbach’s alpha involves division by zero.
Charles
Hi Charles,
So what does it mean if the variance is 0? Can I conclude that the response is highly reliable and consistent then as the variance is 0? Please advice. Thank you.
Sophia,
I am not really sure how to interpret the result in this case. I can see that if I change any single value slightly (e.g. changing 4 to 4.0001), I get a large negative value for Cronbach’s alpha.
Charles
I want to quantify vendor quality rating by using this cronbachs alpha method considering several parameters on quality delivery price services on likert scale.is this method is effective to calculate the vendor is reliable or not?
I don’t know what you mean by vendor quality, but I don-t see how Cronbach’s alpha can be used to quantify vendor quality based on how I would define the term.
Charles
Thanks for this; very helpful. I have calculated alpha using the ANOVA w/o replication model but I need help understanding your equation alpha = 1 – (MSe / MSb). What does MSrows (MSb) represent? I understand that MSe is the mean squared error for the sample? (Sorry for the novice question.)
You can find the definition of MSB and MSE at the following webpage:
Two Factor ANOVA without Replication
Charles
Hello Charles, I have a construct of 5 items in which one item is negatively worded. I recoded the negative item. However when i run my reliability analysis thru spss, I got the Cronbach Alpha of .275 after I reverse the negative score. If I run all 5 items without the reverse score, I got Cronbach Alpha of .604. What should I do?
Naz,
If you recoded the negative item properly, then you should accept .275 as the correct value for Cronbach’s alpha.
Charles
Hi Charles, Firstly I would really like to thank you for responses above as they are of great help for beginners.
Now my query. I have two scales of 30 statements each with FOUR constrcts in first scale and only one constrct in second sacle. Do I need to calculate cronbach alpha value for each constrct. What should I do with negative statement s in each scale
Ripuhaman,
Sorry, but I don’t understand what you mean by “constricts”. Perhaps by four constricts you mean a Likert scale of form 1,2,3,4, but in this case I don’t know what you mean by “only one constrict”.
Also, if by “negative statements” you mean “reverse coding”, then this is explained on the referenced webpage.
Charles
By constructs I mean factors which are further explained in the likert scale by various statements ( items).
Ripudaman,
You need to calculate a separate Cronbach’s alpha for each construct.
Charles
Hi Charles,
Thanks for the reply. Now I have calculated the alpha value . Both ways separately for each constrct as well as for complex scale values are .90 for scale n for each constrct it is .76,.70,.68,.61 respectively . Are these values ok and acceptable. Should I go by the scale . ? Plz suggest what is to be done.
Ripudaman,
There is no agreed upon criteria for acceptable values of Cronbach’s alpha. Some would say that a value higher than .7 is acceptable; some would accept slightly lower values.
I don’t know what you mean by “Should I go by the scale?”
Charles
Hi Charles,
I’ve got an issue with my Cronbach’s Alpha which is negative !!
I do not know where is the problem as it can’t be a calculation error because I did it with SPSS… If you have any idea keep me informed
Thanks
Romain,
It is entirely possible for Cronbach’s alpha to be negative. This shows very poor interval consistency.
Charles
while doing cronbach alpha test for reliability,do questions regarding to all factors can be tested in one test?
You should calculate separate Cronbach alphas for each factor. Obviously you can calculate one value, but it won’t provide a good measurement of the questions.
Charles
I got these results.
Reliability Statistics
Cronbach’s Alpha N of Items
.420 25
What do you think?
That is a pretty low value for Cronbach’s alpha.
Charles
I’ve deleted 8 items and i ve got 0.703.
Is this acceptable? and is it ok to delete some items? thanks in advance.
Yes, it is ok to delete these items, but you need to decide why deleting these items increased Cronbach’s alpha.
It is possible that some of these items were poorly worded or used reverse coding (as explained on the referenced webpage), but it is also possible that these items were on a different topic (or more than one other topic). In fact, it would be interesting to calculate Cronbach’s alpha just on the 8 items that you removed.
Charles
That’s what i got
Reliability Statistics
Cronbach’s Alphaa
-.143
N of Items
8
??
Hi…so say if i have a 5 likert scale assessment for 15 questions, i would need atleast 70 to 80 repondents for cronbach 0.70?
Jamuna,
How did you calculate this? I don’t see how you can guarantee a Cronbach’s alpha of .70.
Charles
Hi ! What if I have 2 questionnaires? Should, I’ll compare the Cronbach’s Alpha of both? And is the Cronbach’s Alpha is the percent of the total number of people?
Is the variance under the questions the percent of the people?
Rin,
You should create two separate Cronbach’s alphas. You don’t need to compare these alphas to each other.
Alpha is not a percentage of the number of people nor is the variance.
Charles
Hi Charles,
Can I use Cronbach’s alpha test to test reliability of a tool wherein responses are simple
“low, moderate, high a” and not ranked on Likert scale
Thanks
Pooja
Pooja,
Essentially this is a Likert ranking, e.g.: 1 = low, 2 = moderate, 3 = high
Charles
Hi,
first of all thank you for the nice wrap up of Cronbachs Alpha, it did help me a lot! I used your way to calculate it in Excel for my Master Thesis.
However, while it did totally work fine for 3 scales, the fourth scale has an alpha of -6.5. I did rescale “negative” questions (e.g. -2 to 2 in case of “i don’t like..”).
Could you please help me out, which reasons are there for a negative alpha? Thanks a ton!
Erik,
If you send me an Excel file with your data and calculations, I will try to figure out what has gone wrong.
You can find my email address at Contact Us.
Charles
Is .096 acceptable or not?
Wesley,
.096 is quite a low value for Cronbach’s alpha and would generally not be viewed as acceptable (unless you meant 0.96, which is just the opposite).
Charles
I am getting an alpha of -1.6. I didn’t think that was possible. Any ideas on what errors I might have made? (I’ve checked it a few times now already). I used the calculation method in Excel that you describe in the beginning (not reliability stats features).
Mona,
You can get negative results, but I can’t remember now whether such a high negative value is even possible.
In any case, without seeing your data and analysis, I can’t say for sure why you are getting such a result.
I suggest that you install the Real Statistics software and use the Reliability data analysis tool.
Charles
ok, thanks. How does one interpret negative results?
Mona,
A negative result indicates poor internal consistency. Potential causes for this are described on the referenced webpage.
Charles
Dear Charles,
I have one problem which I need your support. I made one survey which has 6 factors, each factors I designed two items(questions), actually the designed questions I referred to a matured questionnaire but with adaptation for my research purpose. But I did test for my questionnaire (invited only two friends to participate the questionnaire), but unfortunately alpha value calculated for 5 factors was 0, and only one alpha value is about 0,8.
I really do not understand what is the problem?
1)Is it because tool less test participants?
2)Is it because only two questions for one factor?
3)Or is it because the questions of each factor are too similar?
For example, one question is “do you feel joyful at work?” and second one is “do you feel positive at work?” But actually these questions are used and tested by some other questionnaires. I really confused.
I calculated Alpha by SPSS software. But I will try to use your software to do it once more.
Anyway, I hope you could give me some hint about this problem.
I really appreciate your support.
Best regards
Victor
Victor,
You probably can’t expect too much if you only had two participants.
Using similar questions should increase the value of Cronbach’s alpha.
Charles
Hi,
My variances is all zero since all respondents for my pilot study answered 1 (I am using yes/no type answers). I have 5 respondents. How can I fix this? My alpha given this #DIV/0!
My questionnaire is more like a checklist (to check whether my prototype is designed accordingly or not).
Siti,
There is nothing to fix; with all 1’s the variance will be 0 and therefore Cronbach’s alpha will be undefined. If you are happy with the checklist you don’t really need to perform Cronbach’s alpha on it; if not then all the answers being the same tells you there is a problem.
Charles
Excellent! Charles, for your good job!
I need to know about the Factor Loading,
Could you plz. kindly explain how to find Factor Loading for each item in a questionnaire…
Thanks in advance!
Venkat,
See Basic Concepts of Factor Analysis
Charles
can split half reliablity of a tool be greater than crohnbach’s alpha?
Riya,
I have not investigated this, but probably it is possible.
Charles