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 two halves and compare the results.
Topics:
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 two halves and compare the results.
Topics:
How can I calculate the sample size needed for reliability analyses for a 60 item knowledge measure?
Janice,
The various reliability tools described on the website produce a reliability measurement. This not the same as a test and so there is no sample size requirement (although most require a sample of at least 2 or 3 in order to successfully calculate the measurement). If, however, you want to evaluate this measurement in some way, then it can be appropriate to ask about the minimum sample size. There are two commonly used types of evaluations: (1) is the reliability measurement significantly different from some value (often zero) or at least greater than some value and (2) what is the confidence interval for the measurement.
Which reliability measurement do you plan to use and how do you want to evaluate this measurement?
Charles
Hi, I have read the literature, and most of the reliability checks are done on the questionnaires. How if my data is not a questionnaire? How should I split the data into two halves? Should I sample with or without replacement?
Thanks in advance :>
Hello Pante,
The data doesn’t need to come from a questionnaire to use the split-half method.
Usually, the split is using a predetermined approach: odd-numbered items vs even-numbered items or first half of the items vs second half of the items. You could also take all possible splits and then take some sort of average of the results. Cronbach’s alpha and Guttman’s reliability are two approaches for doing something like this. All these are described on the Real Statistics website.
Charles
Hi, can I use split half for reability that measure personality based on the MBTI? For example I have four constructs, one of them is Introvert/Extrovert.
The scale I use is semantic scale (6 points), one side Introvert and one side Extrovert.
I am not familiar with the MBTI nor your specific questionnaire and so I cannot answer your question without more information.
Charles
Hi Charles
I want to know how to judge the level of Split-half Reliability（Spearman-Brown，Guttman Split-Half）,which specific coefficient values represents poor（unacceptable）；slight agreement；moderate(fair)；good；excellent? Thank you very much.
Hello Keke,
There isn’t universal agreement about this, but you can use the guidelines given for Cronbach’s alpha. See
Cronbach’s alpha
Charles
Hi Charles
I want to know how to judge the level of Split-half Reliability（Spearman-Brown，Guttman Split-Half）,which specific coefficient values represents poor（unacceptable）；slight agreement；moderate(fair)；good；excellent?
In addition，McDowell showed that typical Pearson correlations for interrater reliability in the scales in the range 0.65 to 0.95, and values above 0.85 may be considered acceptable.Pearson correlations for repeatability are often high,falling between 0.85 and 0.90.But no detailed rating criteria，do you know the detailed rating criteria of Equivalent-form Reliability and Scorer Reliability? Thank you.
Keke,
These values seem a bit high for reliability. See my response to your later comment.
Charles
Sir ,
I am a college student and I was assigned to Split half method using spearman rank order correlation coefficient. Hmm. In split half method what does it mean by spliting two halves into odd-even items?
If my question Set A (multiple choice for 15 points) and Set B (identification 20 points ) ? It is correct or not?
Hello Anilyn,
Odd questions are numbered 1, 3, 5, … and even are 2, 4, 6, …
You can only calculate a coefficient for questions that are measuring the same concept. If Set A and set B are measuring different concepts then you need to calculate separate coefficients, one for each set. If you set of questions includes identification items such as Gender, City, Height, Level of Education, etc., then it doesn’t really make sense to calculate a coefficient for these questions.
Charles
How do I get x and y scores on 160 questionnaires administered in different schools
Hello Andrew,
Sorry, but I don’t understand your question. Are you asking whether Split Half can be applied to questionnaires from different schools?
Charles
Hello sir, is it okay to have two kinds of scale in my whole questionnaire? I have 52 questions and 50 questions has likert scale and 2 questions has guttman scale. If possible, do I have to calculate the reliability and validation differently between likert and guttman scale?
Hello Boddhi,
1. If the two questions using the Guttman scale are testing a different concept from the other 50 questions, then know you should not use Cronbach’s alpha for all 52 questions. This is because separate Cronbach alphas should be calculated for questions that are testing different concepts.
2. I have never tried to use Cronbach’s alpha with questions that use Guttman’s scale and so I don’t actually know if this is possible (even when all the questions use Guttman’s scale. Clearly test-retest should work in this case.
Charles
hello sir.
I want to test my questionnaire with splihalf method and it consists 60 questions but i am not getting that how can I give scores to the questions.Please tell me the step by step procedure of this method.
What sort of questions do you have? Multiple choice? Questions that have a correct answer and one or more incorrect answers?
Charles
Sir
I have some multiple choice questions and some open ended questions thats why I am confusr to give scores.
Thanks &Regards
Hello Akriti,
If all the questions have a correct answer, then you can use the values of 1 for correct and 0 for incorrect.
If the questions are testing different concepts then you need to conduct different split-half tests for questions of each type (where a pe refers to a concept).
Charles
Hi Charles, I am tasked to make a Split Half Test, can I make a split half test with a positive and negative statements? And if yes so, how?
Becca,
Yes, the split half method can handle both positive and negative statements. How this is done depends on what the statements are measuring.
E.g. if there are correct answers and wrong answers, you can code 1 for the correct answers and 0 for the wrong answers. So the statements “Russia is bigger than Germany” (positive statement) and “Germany is not bigger than Russia” (negative statement) are both given the code 1 (correct answer).
If instead the statements represent an evaluation, then the important thing is to make the direction of the answers consistent. E.g. suppose you are evaluating whether someone likes sports, the response to the statement “I like basketball” is either True or False. If we decide to code True by 1 and False by 0, then we need to reverse code the statement “I don’t like basketball”; this time True is coded by 0 and False is coded by 1.
Charles
Thnks sir, but I use straight 24 items directly, my supervisor want me to explain how I got 0.7 but I use system to calculate it. Pls how do I do d random shd I skip 18 or 19 pls its urgent I ve gone too far in my research I ve even use wilcox z test for analyr can I still use croubach alpha since my data is likert scale but alpha gave me 0.92 pls reply
Sophia,
I don’t fully understand your comment.
It is very surprising to me that you would get .3 from split half and .92 from Cronbach’s alpha. This is likely to mean that you made a mistake.
Charles
Pls how do I calculate 25 items using split half odd and even. Thnks
Sophia,
One approach is to leave one question out at random. Another is to use Cronbach’s alpha instead of the split half methodology.
Charles
Sir, please can I use split half for an achievement test of 10 items?
Pls I need reply urgently, I ve 25 items and I using split half but the answer is very low less than .3 my questions is in 4 point likert scale wht do I do? Pls
Sophia,
There are a number of possible causes for a low value, including:
1. Poor wording in the questionnaire
2. A questionnaire that is measuring more than one concept
3. Reverse coding of some of the questions
You can get more information about these issues by reviewing the following webpage about Cronbach’s alpha. Cronbach’s alpha is another measure of reliability, but the concepts are pretty much the same.
Cronbach’s alpha
Charles
Hi charles, in my thesis i got scores of cronbach’s alpha ranging between 0.6 and 0.8. My supervisor asked me to explain how i dealt with the errors. kindly explain to me what i can do since am not very good in statistics. Thanks
Esther,
What errors is he/she referring to?
Charles
Hi Charles,
I am wondering if I can use the split halves method when giving partial credit and how this can be done.
Thank you!
Margie,
Yes, the split-half approach handles partial credit. If each question is say worth 10 points, then you can assign any value from 0 to 10. The methodology supports these types of scores (as described at https://real-statistics.com/reliability/split-half-methodology/split-half-basic-concepts/)
Charles
Pls Charles is there any reasons for not conducting pilot study in Mixed methods research
Mohammed,
Given that I don’t know the context for your question about a pilot study in mixed methods research, I don’t know of any reason for not conducting such a study.
Charles
Sir, I constructed a Socio-economic status scale for students which has multiple choice & one word answer type questions. These questions are related to parental education, their income, living style etc. How I do item analysis and how to get its split-half reliability?
Kul,
You can do both of these tasks as described on the website provided you are able to give the answers a score that can be viewed as correct or partially correct. Also see Item Analysis.
Charles
Hi Charles, thanks for the good work. I adapted the WHO STEPS instrument for identification of risk factors of Non-communicable diseases to identify risk factors of essential hypertension via a retrospective hospital based case control study in my locality. the WHO STEPS instrument only seek to identify previous occurrences and their pattern among cases and controls with a combination of YES and NO questions, MULTIPLE CHOICE questions and OPEN ENDED questions which do no have an actual scoring system. I have also not found any previous psychometric properties of the WHO STEPS instrument. please, how do I validate and establish reliability of the adapted instrument
Soupriye,
I am not familiar with the WHO STEPS instrument, and so I would need more information about it before I could help you. Is the problem that you are having the fact that (1) there is a mixture of question types and/or (2) there is no scoring system for the open-ended questions? Or is there some other issue?
Charles
Greetings Charles.
I have 15 question of a test. I want to calculate the reliability of the test.
The first half of the test contains 8 items (1-8) and I calculate the total points from these items. The second half of the test contains 7 items (9-15) and I calculate the total points from the second half. Then I calculate the Sperman Brown formula between these two halves. Is it true that I thought so? Would it be correct to create equivalent halves like 8 and 7? And is there any source that I can refer to?
Thank you very much
Cenova,
That the halves have different sizes is not important in calculating the correlation coefficient. The Spearman Brown correction is slightly affected as explained at https://real-statistics.com/reliability/split-half-methodology/spearman-browns-predicted-reliability/
What is more important is that the first 8 and last 7 questions are testing the same thing. If you are unsure of this, it would be better to use an odd/even split. You can also use Cronbaach’s alpha instead.
Charles
Hi,pls can u tell me the statistics for split half method becus my supervisor said there is a statistic for it and I dnt think I knw it but am guessing it should be the formula but not sure……pls can you tell me the statistics?
The referenced webpages show possible split-half formulas.
Charles
hi sir. i m constructing a scale. its reliabiltiy coeff (odd even) comes to 0.619 and brown prophecy r comes 0.764. wt is ur analysis regarding its reliability. actually it is said that r between .7 to .8 is considered good for the scale. shud i go for the pilot study again r there u think is no need to worry abt….
Sajad,
This seems to me to be a reasonably acceptable result.
Charles
Hi Charles I’m doing a research regarding professional teaching delivery with 4 factors:teaching skills, instructional materials, classroom management, and content knowledge. I have 10 questions for each factor for a total of 40 items. What tool can u advise me to utilize for the reliability of the questionnaire?
Paul,
One approach is to calculate four separate Cronbach’s alpha for each block of 10 questions (for each factor).
Charles
Hi. For construct validity what test to be used. I have used cronbach alpha and the reliability score is good. This is for 18 questions, 5 likert scale on psychological stress. Thank you in advance.
Validity depends on the specific information domain, and so I don’t have an easy answer for you without more information. Sometimes you could show that the results are comparable to another test that has already been shown to be valid.
Charles
I have a 20-point multiple choice test on measures of centrality. Is it appropriate to use the split-half to test its reliability?
Thank you!
Jack,
Yes, you can.
Charles
Thank you Charles!
Why is it appropriate to use the split-half method to test the reliability of my instrument?
Hi again Charles. I sent a note yesterday about calculating SplitHalf reliabilities yielding several that were >1.0. However, I just realized that the ones yielding Spearman-Brown SHs > 1.o were negatively correlated (due to the fact that what I entered were difference scores). I just re-reran the with my own formulae after calculating the absolute-values to make the correlations positive and yielded more reasonable results (i.e., reliabilities <= 1.0).
Thank you again for your wonderful tools.
Bill
Bill,
Good to hear that you were able to resolve the problem yourself.
Glad you like the tools.
Charles
My instrument is 30 question for multiple choice quesion and each scoring for right answer is one Mark and incorrect answer is zero mark can I use split half method and how
Nagar,
Yes, you can certainly use the split-half method. Just mark each question as 0 or 1, as you described and use the approach described on the referenced webpage.
Charles
If I have 15 question of likert scales, can I use the split half method in excel
Thanks
Yes. This is explained on the referenced webpage.
Shortly, I will add some new capabilities, but the current version of the webpage can still be used.
Charles
The first half will be the sum of the 7question and the second half will be the sum of the 8question
I am using your mentioned formula
=CORREL
and got the r =0.56
Can I say it have a moderate reliable result? Or I just got a low reliable result
Thanks for your promply replied.
Hello,
Can you tell me how to use the split-half method in Excel? I don’t see an option for that method. I just need to know the steps to use the function in Excel.
Thanks,
Autumn,
Excel doesn’t have a function of data analysis tool to perform the split-half method.
The Real Statistics add-in to Excel provides the SPLITHALF function to perform this method. Alternatively, you can use the Reliability data analysis tool and choose the Cronbach’s Alpha option. This will generate two forms of the split-half method.
Charles
Hi.
I am currently translating a questionanaire from english version to my native language.
the original English version was tested using the Guttman Split Half for its reliability.
Therefore for the translated version, do I have to use the same reliability test or can I use Cronbach’s alpha to test for its reliability?
Thank you
Nicolette,
You don’t have to use the same reliability test.
Charles
hey can you please explain me how to calculate t-value from percentile rank. And how to infer Cronbach alpha from split half reliability.I am not able to catch up stats can you help?
Arpana,
I don’t know how to calculate a t value form percentile rank nor how to infer Cronbach’s alpha from Split-half reliability.
Charles
Thanks Charles.
May I know what is the difference between Spearman and Guttman split half?
Thanks in advance
Please i dont understand how the correlation coefficient came about… U only showed the formula and the answer. Please show the working let it guide me for my research paper
Nelly,
See Basic Concepts of the Correlation Coefficient.
Charles
Hi
In a questionnaire based study reliability of the instrument is essential. I have a study about knowledge of vaccine preventable disease among nursing students. Nursing students require to take certain vaccines to protect themselves from infection. we selected 8 such diseases. The questionnaire assessed
1. how many students can correctly identify the selected 8 diseases as vaccine preventable (ie. knowledge). we make a sum scale of knowledge 0 to 8 ( 1 for correct response and zero for incorrect
2. how many students has received advise to take vaccines and from which source
comprising of 13 questions :
A) 4 on socio-demographic data (age, gender, monthly income, level of study ), they cannot be given 1 or zero score for response.
B) 8 closed ended questions (yes /no) about identification of 8 diseases – they can be scored 1 for yes and zero for no
c) 1 close ended questions on source of advise to take vaccine ( paper / internet/ friend) . the responses can not be numbered like previous yes/no close ended question
my question is can i use half spilt theory to evaluate reliability of this questionnaire?
i cannot give score for every answers made in sociodemographic section (ie they cannot be scaled)
similar thing is applicable to source of advise section. frequency of different sources can be estimated but they cannot be scaled like knowledge score.
its only the knowledge scale part which can be put under test by HS method.
My question is in this type questionnaire does reliability depends only on the scaled measures (e.g knowledge scale) and not on unscalable measures (e.g. demographic data, source of knowledge)
It is unlikely that you will accomplish anything by performing reliability testing on demographic questions. You could use split-half or Cronbach’s alpha on knowledge questions.
Charles
Thank you for your software. I can not get the split-half methodology to work for me. I’m unclear if I’m suppose to use one of the functions and type in =SPLIT_HALF(numbers). I cannot find split half on any of the functions. Could you please help me with this. Thank you.
I figured it out. Thanks anyways!
how did you do it?
I have a utilization of health services questionnaire. It has 42 questions. some questions have multiple right answers , then how do i do the scoring for split half reliability test?
Kamlesh,
You can score each question by 1 for a correct answer and 0 for an incorrect answer.
Charles
I have a 98-item multiple choice test that I want to pilot test. However, the school just gave an hour for the testing. Is it ok if I split the test into two and give half to one class and the other half to another class? If this can be done, how do I compute for the reliability of the entire test/instrument. Thanks.
Eleanor,
The problem with this approach is that you may be testing the difference between the two classes rather than the internal consistency.
Charles
Pls, can i use split half methodology on multiple choice questionnaire, if yes how will i group them.
Wale,
You can use the split-half method for multiple choice questions. Use the coding 1 for the correct response and 0 for an incorrect response. You can split odd numbered questions versus even numbered questions.
Charles
thanks for the info.
what are the steps in computing coefficient of split-half reliability ?
Nithin,
I believe that these are stated in the referenced webpage. Please explain what you mean.
Charles
if split half reliablity coefficient is 0.67 then is it reliable?
Riya,
This is a judgement call. Some would say that the value needs to be at least .7. There isn’t universal agreement about what is an acceptable value.
Charles
Dear Sir,
I am doing a research on the feasibility of opening a child care centre. My survey questionnaire include 10 questions on parents preference in choosing a child care centre. Some are ‘yes’ ‘no’ questions and the rest with maximum options. Can you please advice me whether I can test the reliability of the questionnaire and if yes, how?
Sajitha,
I’m sorry, but what are questions with “maximum options”?
Charles