Split-Half Methodology

One way to test the reliability of a test is to repeat the test. This is not always possible. Another approach, which is applicable to questionnaires, is to divide the test into even and odd questions and compare the results.

Example 1: 12 students take a test with 50 questions. For each student the total score is recorded along with the sum of the scores for the even questions and the sum of the scores for the odd question as shown in Figure 1. Determine whether the test is reliable by using the split-half methodology.

Split-half methodology

Figure 1 – Split-half methodology for Example 1

The statistical test consists of looking at the correlation coefficient (cell G3 of Figure 1). If it is high then the questionnaire is considered to be reliable.

r = CORREL(C4:C15,D4:D15) = 0.667277

One problem with the split-half reliability coefficient is that since only half the number of items is used the reliability coefficient is reduced.  To get a better estimate of the reliability of the full test, we apply the Spearman-Brown correction, namely:

Spearman-Brown correction

This result shows that the test is quite reliable.

Real Statistics Functions: The Real Statistics Resource Pack contains the following supplemental functions:

SPLIT_HALF(R1, R2) = split half coefficient (after Spearman-Brown correction) for data in ranges R1 and R2

SPLITHALF(R1, type) = split-half measure for or the scores in the first half of the items in R1 vs. the second half of the items if type = 0 and the odd items in R1 vs. the even items if type = 1.

The SPLIT_HALF function ignores any empty cells and cells with non-numeric values. This is no so for the SPLITHALF function.

For Example 1, SPLIT_HALF(C4:C15, D4:D15) = .800439.

Example 2: Calculate the split half coefficient of the ten question questionnaire using a Likert scale (1 to 7) given to 15 people whose results are shown in Figure 2.

Split-half data

Figure 2 – Data for Example 2

We first split the questions into the two halves: Q1-Q5 and Q6-Q10, as shown in Figure 3.

Split-half test

Figure 3 – Split-half coefficient (Q1-Q5 v. Q6-Q10)

E.g. the formula in cell B23 is =SUM(B4:F4) and the formula in cell C23 is =SUM(G4:K4). The coefficient 0.64451 (cell H24) can be calculated as in Example 1. Alternatively, the coefficient  can be calculated by the worksheet formula =SPLIT_HALF(B23,B37,C23:C37) or =SPLITHALF(B4:K18,0).

We can also split the questionnaire into odd and even questions, as shown in Figure 4.

Split-half odd-even

Figure 4 – Split-half coefficient (odd v. even)

E.g. the formula in cell L23 is =B4+D4+F4+H4+J4 and the formula in cell M23 is =C4+E4+G4+I4+K4. The coefficient 0.698813 (cell R24) can be calculated as in Example 1. Alternatively, the coefficient  can be calculated by the supplemental formula =SPLIT_HALF(L23,L37,M23:M37) or =SPLITHALF(B4:K18,1).

99 Responses to Split-Half Methodology

  1. Ali says:

    I will like to know the conditions that calls for the use of split-half with likert-scale or ordinal data.

    Best Regards

    • Charles says:

      I don’t know of any conditions, except that we are assuming that the distances between adjacent items in the scale are equal; i.e. if you are using 5 scale Likert (1, 2, 3, 4, 5), that the distance between 2 and 1 is the same as the distance between 4 and 3, etc.

  2. Nemat says:

    Dear Dr. Charles
    Can I use split-Split-Half for 5 point Likert or ordinal scale?

    Best Regards

  3. Bec says:

    Many thanks Charles for a fabulous explanation!
    Once we have calculated our split-half (and in my case applied the Spearman-Brown Prophecy correction), how do you suggest we interpret the findings?
    Is there a universally agreed set of categories for high/mod/low correlation for the split-half? such as those put forward by Landis & Kotch (1977) for agreement analyses.
    many thanks for your help,

    • Charles says:

      From what I can see from the literature, there isn’t universal agreement even about the interpretation of Landis & Kotch (1977). The best I can say is to use the guidelines for Cronbach’s alpha (see Cronbach’s Alpha), but again caution is advised since there isn’t universal agreement about these guidelines.

  4. Abbass says:

    Hello Sir,
    I wonder if the Split-Half Methodology applies on school exams (Multiple choice assessments) to test its reliability

    thank you

    • Charles says:

      Hello Abbass,
      Yes, you can use the split-half method (or Cronbach’s alpha) in this case. Remember to code the answer as 1 = correct answer and 0 = incorrect answer.

  5. Lindah says:

    hi,I am doing an exam tomorrow, but I am
    trying to understand the split half method but I’m not getting it,after splitting into half’s how do you find your correlation coefficient, help me please

    • Charles says:

      You just calculate the correlation coefficient between the two halves. In Excel this is done using the CORREL function.

  6. Amy says:

    Does the coefficient found give any information about the validity of the test?

    Thank you

  7. Hassan says:

    I want to do split-half method to test the internal consistency reliability of a new questionnaire, which contains 6 domains, and each domain include different questions; for example, one domain includes 1 question while other domain includes 10 questions. So may question is which is better, getting the average for each domain then doing split-half reliability for these 6 domains as 6 questions or splitting the questionnaire according to the actual questions (in this case how can deal with if the one domian contains more than questions and one of them has sub- divisions, i.e., 2a, 2b,23……
    Kind regards,

    • Charles says:

      I suggest that you calculate a separate split-half (or Cronbach’s alpha) measure for the questions that relate to each of the domains. This approach won’t be very useful for the domain that only has one question.

  8. Rona says:


    Is the split test still applicable to a test it lacks precision and the questions do not coincide with each other?

  9. Niels says:

    Hi Charles,

    Half of my items should be reverse-scored before doing the analysis. Does your set of macros allow for reverse-scoring multiple items?



    • Charles says:

      No, you will need to do that manually.

      • Niels says:

        Hi Charles,

        Thanks for your reaction. I have a different question. Whenever I use one of your Reliability Procedures I get an error saying:

        Cutoff must be a number between 0 and .5

        It has a standard value of “0.27”. Only when I change the point into a comma I get things to work. Is this an issue with my version of Excel, or a more general problem?



        PS I am using your functions for didactic purposes in a methodology course

        • Charles says:

          For some languages comma is used for the decimal symbol, while others use the period. Everything works fine in either case, except when I provide a decimal such as 0.27 as a default. I have tried to fix this problem many times, but I still haven’t come up with the right solution. Fortunately the workaround is simple, namely you just need to re-enter the value, as you have done.

  10. Babe says:

    Sir, doing a comparative study between two groups using the same tool. Want to calculate reliability. I plan to use split half method. My question is… Should I split the tool into two for each group or can I use the two different groups as two halves for the same tool?

    • Charles says:

      I would need more information to give a definitive answer, but most likely you need to make two separate tests, one for each group.

  11. Lira says:

    In one dimension I have only 3 items. Should I calcualte split-half coef. or are 3 items not enough? Well, SPSS does calculate it, but I’m note sure if it makes any sense to calculate it…

    Thank you very much!

    • Charles says:

      You certainly can calculate it, but I would guess that confidence interval for answer is quite wide, probably making the result not very useful.

  12. Andreea says:

    Dear Charles,

    I am testing (as a uni assignment) the reliability of PRT Scale (questionnaire) and I don’t quite understand what requirements does your data need to meet before you can run a split-half. The scale originally had 25 questions (items) but we excluded one due to not meeting the .7 Cronbach’s Alpha value. I should perhaps mention that the questionnaire is in 3 different subscales with less than 10 items in each. Hope this makes sense.

    Thank you!

    Best regards

    • Charles says:

      Dear Andreea,
      I can’t think of any assumptions for using the split half reliability, except to make sure that the questions are split in some way (e.g. odd-even) which doesn’t bias the result.
      Also the questions need to test the same concept; otherwise you need to run separate split-half analyses for each concept (i.e. for the questions that relate to each concept). If the subscales that you refer to are for different concepts, then you will need to run three different split-half analyses.
      One last point. The number of questions being measured also has an impact on the reliability score. See, for example, the website http://blog.questionmark.com/how-many-questions-do-i-need-on-my-assessment.

  13. mubarak ibrahim says:

    Very interested and helpful text. Thanks alot

  14. FATEMEH says:

    Hello dear Charles,

    I have constructed a tool to be used in my Ph.D project, but as am new in the statistical field so am confused and need help. I was hoping if you could help me out with the methods and calculations! I,m not sure if i am doing mine correctly! I appreciate if you don,t mind to guide me.


  15. Armelle says:

    Dear Sir,

    I have an exercise: an algebra test was given to 20 students and for the ten items of the test the answers are 1 or 0. A split-half procedure was conducted. Will I use Spearman-Brown as an alternate method for estimating reliability or KR20 (because it is a dichotomous test)?

    • Charles says:

      You can use either the split half method (with Spearman-Brown) or KR20 to test for reliability. As you said, you can use KR20 since the data is dichotomous. Also since the data is dichotomous, KR20 is equivalent to Cronbach’s alpha.

  16. cj says:

    what is the significant level of spearman Brown, at what point can i say that the test is reliable or unreliable

    • Charles says:

      There is no universally accepted answer to this question, although you will most commonly see that a value of .7 or higher is acceptable, although sometimes you will see that .6 or higher is acceptable.

  17. C J says:

    Goodmorning sir.
    Pls i want to know the appropriate ways of reporting different reliability analysis, eg split half, cronbach, f text, etc, and wen my result is significant

    • Charles says:

      If your result is clearly high enough (see my answer to your other comment), you can say the following

      The XYZ questionnaire has a high level of internal consistency based on a Cronbach’s alpha of .85.

      If the value is low, say .40, then say it has a “low” or “unacceptable level …” With a value such as .65, then I would probably say something like

      The XYZ questionnaire has internal consistency based on Cronbach’s alpha of .65.

      With Cronbach’s alpha, I would also report on what the value is when each of the questions are removed

      … when question #7 is removed then Cronbach’s alpha increases to .75, which is the greatest increase in Cronbach’s alpha based on the removal of one question.


  18. Helen says:

    sir, is focus group discussion and in depth interview a rich instrument for data collection on rural farmers information needs and seeking behaviour towards climate change adaptation. thank

  19. venki says:

    Hi, what is the minimum number of respondents require for checking the reliability in split-half method

  20. Pree says:

    For my thesis I have section two sections, first is multiple choice questionnaire and second is 5 point likert scale. Can I use split half for Mcq and crohnbach alpha for likert scale? Is it right?
    Plz reply

  21. Jen says:

    Hi Charles,
    I’m totally new to the field of statistics and still puzzled….
    In my data sample Spearman Brown prophecy = 0.35, and KR20=0.735.
    KR20 indicates good reliability, the value for SB is rather low. What does it mean? Why are the values so different?

    Thank you!

  22. Unwana says:

    Tnks so much for this…my question is ,how do I go about testing for my 31 question questionnaire’s reliability , do I go ahead and administer to my sample of 110 or do I first administer them to some members. Answer asap please.

    • Charles says:

      Either is possible. Sometimes you administer the test to a smaller sample to check whether the reliability is good. In this way you can change the questionnaire before giving it to the full sample.

  23. Candice says:

    Hi there,

    I was hoping you could help me with my calculations! Im not sure if im doing mine 100% correctly! ;-( Do you have an email address?

  24. Fatin says:

    There are many ways to determine reliability as internal consistency, repeat test, split-half,….. .
    Do the reliability coefficient are the same with all?

  25. Praise says:

    after splitting the items into 2. can we use pearson r to get d correlation coefficient before stepping it up with spearman brown correction formula? thank u.

  26. MEJA says:

    hi, i have the problem on finding the spearman brown when the correlation coefficient is negative number

    • Charles says:

      If your correlation coefficient is negative, then you already know that reliability is very low and so you really don’t need to bother with spearman brown.
      Using spearman brown without any sort of correction will result in a correlation less than -1, which is not very desirable. I would guess that the correct calculation is to apply spearman brown to the absolute value of the correlation coefficient and then negate the result, but I have not seen this done before.

  27. jane robson says:

    I am analyzing data from subjective scoresheets (10 questions regarding performance and scoring against performance). I am feeling that the questions are subjective and being scored differently by different scorers. How can I measure the reliability. I used Spearman Brown (.065)& Pearsons (.12) but am questioning this. Please assist

    • Charles says:

      Please describe the scoring used in a little more detail. Are people rating some characteristics using a Likert scale (say 1 to 5) or is this a typical multiple choice question test or something else?

  28. Binu says:

    Dear Sir,

    I would like assess the reliability of a questionnaire in an experimental study. I have done pre and post test among 15 samples as my main study will be 150. How do I calculate reliability?

    • Charles says:

      The usual approach is to use Cronbach’s Alpha. Although this measurement has a number of flaws, it is the commonly used measurement.

  29. Denise says:

    Thank You! :) big help. I have 2 questions though. What if the correlation coefficient’s equation was (0/0), is that considered to be 0.00 for that matter? What is the interpretation of a 0.00 for correlation coefficient and spearman-brown correction? Thank You!

    • Charles says:

      A correlation coefficient of zero means there is no correlation, which in this case indicates very poor reliability. Spearman-Brown is undefined (can’t divide by zero), but there is no need to calculate it since the zero value tells you all you need to know.

  30. Alam says:

    Dear sir,
    Do I need to test the reliability of a likert-type 5 point scale with the data from a pilot study before applying the instrument for the final study? I am new in statistics please answer.


    • Charles says:

      You don’t have to do this, but it is commonly done.

      • Alam says:

        Dear Sir,
        If I use without testing the reliability of the instrument and it is discovered during analysis that the instrument was not reliable then I have to do the research again. That is my understanding, I might be wrong. That why is it not right to check the reliability of the instrument through a pilot study?

        I am expecting your clarification of the matter so that, I can get a clear idea.


        • Charles says:

          It is best to find out if there are problems early (e.g. with a pilot with say 30 subjects) rather than for the full experiment (e.g. after you have given the questionnaire to say 2,000 people). If reliability is a problem you can fix the questionnaire and then repeat the pilot. If you only check reliability after giving the questionnaire to 2,000 people you will need to repeat the questionnaire with all 2,000 people if you find there is a problem.

  31. claudette phoenix says:

    Thank you. The information is extremely helpful. Can I use the split half methodology with a test that has two sections – Multiple Choice and Essay? Can I, instead of using odd vs even , use multiple choice vs essay?

    • Charles says:

      The idea of the Split-Half Method is to make a random split of the test items. Odd/even is just one way to do this. However, multiple choice vs essay may not be a random split, and so I would be a bit cautious of using such a split.

  32. Meera says:

    Dear Sir,

    I’ve been searching for some one who could explain Cronbach’s alpha and Spearman-Brown split-half. You have done a wonderful job. Thanks a million. I have a doubt though. Plz do forgive me if itz undiluted stupidity (I’m new to Statistics). Do they (Cronbach’s alpha and Spearman-Brown split-half) test both reliability and validity or just one and not the other? I’ve seen a study where the researcher used Spearman-Brown split-half to test reliability and Cronbach’s alpha for validity. Hence the question.


  33. Thanks a lot. It’s a very nice presentation for the researcher. May Allah bless u. You may present all of the methods of reliability and validity test with formula and indications for excel or SPSS. You may also present directions regarding test of hypotheses.
    best regards

  34. GYPSA K P says:

    can we use split half for 15 item tool

  35. Karisa Walker says:

    I have another question…So if I am giving my survey to 430 students, do I need to conduct the split half test on a small number of the sample to see if its reliable before I give it to all of the students?

  36. Karisa Walker says:

    Great information! Is it possible to get a reliable score using the split half method with just a 10 questions survey?

    • Charles says:

      Yes, you will get a reliable score for the split half method with just 10 questions. The real question is whether the 10 question survey is reliable (i.e. internally consistent). I don’t have a simple answer for this question. All I can say is that, in general, more questions increases reliability. However, if you add a question which is poorly worded, tricky and/or testing something different from the other questions, then reliability can decrease.

      • Karisa Walker says:

        Thanks for the response. It’s a student survey for elementary students so I was worried that more questions would prompt false responses. Decisions….Decisions….

  37. ajecus says:

    sir, is this a tool used for item analysis? the reliability of an item?
    asap reply is much appreciated.

  38. Ratul dad says:

    Thank u sir. I have been benifited. I m ever heart to u sir

  39. Iqra Aazmat says:

    Thank you very much for giving very much clear concept! i got the idea about split- half reliability.

  40. JISHA says:


    R1 : 1, 8, 3, 7, 5, 10, 4, 6, 2, 9
    R2 : 1, 8, 4, 7, 5, 9, 3, 6, 2, 10

  41. Dwipta Dey says:

    Thanks for this clear and concise explanation. Can I use the rank order method to calculate the r ? Is it good enough to establish the reliability of a structured knowledge questionnaire?

    • Charles says:

      I would guess that it depends on the motivation you have for using the rank order correlation (i.e. Spearman’s rho) instead of the Pearson’s r.

  42. Tjang Kian Liong says:

    Consice and clear. It is a great help for me. Thank you very much.

  43. Groover says:

    Hi Charles,
    Any thoughts how to calculate Spearmann-brown for Likert 5 point scale answers? How to split the answers?

    Thank you very much

    • Charles says:

      The calculation is exactly the same as described in the referenced page. You can split the questions by odd-even as described on the referenced webpage or first half of the questions vs. second half of the questions. It is important to decide on which approach to use in advance and not after you have seen the results (which are likely to be different); alternatively you can report on both approaches.

  44. Parto says:

    Hi, Thank you very much for a very helpful article. I have downloaded The Real Statistics Resource Pack to my Excel. However, it doesn`t have SPLIT_HALF function. Any suggestions?
    Thanks and kind regards

  45. Jairo says:

    Big help, thank you!

Leave a Reply

Your email address will not be published. Required fields are marked *