Dear Dendy,

I think its not matter of number of questionnaire but few question are important for it - What data you have collected? what is purpose? How you are doing analysis of this data? As per your research question, objectives and Hypothesis you have to identify variables for analysis.  So depending up on the statistical test what you are selecting  you have to think the scale. for example If you are using Bi-Nominal test you have to convert it in to 3 point scale. So its depends exactly upon what data analysis test you are using to analysis of data and interpretation of the data for your Hypothesis testing. So you have to talk with your research supervisor for the further processing of the data.

I hope this would be helpful !

Best wishes...~

Take important parameter in your questionnaire use 3 point or five point likert scale for all parameters.

Hello Dendy.

Can you please try to make your scale converted to binary (0-1). You can then use any types of likert scale as you want.

Dear Dendy,

Firstly you need to think whether scale conversion has an impact on your measurement. Simply you can convert, for example, 5 point scale to 7 point scale or 7 point scale to 5 point scale as follows.

5 point scale to 7 point scale

1 ≈ 1

2 ≈ 3

3 ≈ 4

4 ≈ 6

5 ≈ 7

7 point scale to 5 point scale

1 ≈ 1

2 ≈ 1

3 ≈ 2

4 ≈ 3

5 ≈ 4

6 ≈ 4

7 ≈ 5

Good Luck!

Are you trying to combine them into one scale/index?  However, think very carefully about doing this.  Why did you use a 7 point scale for some items and a 5 point scale for others?  How will changing the scaling after data collection affect  your rationale?

If not, leave them as is.  Depending on how the data looks you may want to collapse categories for reporting and/or certain analyses so that you have a sufficient number of respondents in each category.  For example you may collapse strongly agree and somewhat agree into agree and strongly disagree and somewhat disagree into disagree.  This would give you three categories of agree, neutral and disagree.

I agree with the answer from Joyce Morris. In general, you should keep and use the full amount of information you collected, so retain the 7-point scale as is, for most analyses. 

But to simplify your reporting (and depending on how comfortable the readers of your report are with numerical detail), you might want to collapse both the 5 point and 7 point scales down to just 3, or maybe even just two for reporting purposes. Two categories (e.g., any agreement vs neutral or disagree) has the reporting advantage that a single number summarizes each item -- "% agreeing".  

Unless your total sample is very large, it might also be necessary  to collapse down to fewer categories in cross-tabulations, to meet the cell size requirements for chi-square.

However, if you want to combine the items into a single index, then the difference between the 7-point and 5-point scales does matter. If you don't make the scales consistent (especially, by giving them similar dispersion), then implicitly your index will be giving about 40% more weight to the item with a maximum score of 7 than to an item with maximum score of 5.  

There are at least two ways to fix that issue.  The most common way is to convert all items to z-scores, where the z score for each respondent is z = (respondent's item score minus item mean) divided by (item standard deviation).  The resulting z-scored versions of the items will all have an s.d. of 1.0, and so they all will get equal weight when summed to form the index.  The resulting index will take on both positive and negative values, with a mean close to zero, but you can shift its scale to be all positive if you want, with any minimum/maximum you find convenient (say, 1 to 3 or 0 to 100).

A statistical purist might not be comfortable with the z-score solution,  It requires using the mean and s.d. of Likert-scaled items. Those items are (strictly speaking) only ordinal measures whereas the mean and s.d. presume interval measurement.  If that issue rules out z-scores for you, you can do approximately the same thing by just recoding the items so that they have the same range. I prefer to leave the 7-point scale as is, and expand the range of the 5-point scales to match, as follows:

1 = 1

2 = 2.5

3 = 4

4= 5.5

5 = 7

That recode increases the range, while maintaining equal distances between the 5 categories. (Terans' suggested recode  from 5 points to 7  introduces unequal distances, without any obvious justification.) Maybe the "true" distances aren't equal, but you have no way to know what the "true" distances really are, so the assumption that they are equal may be more defensible than assuming any other, arbitrarily selected, set of differing distances.

Alternatively, you could contract the range of the 7-point item, but then it's a little trickier to maintain equal distances between the 7 categories. Terans' suggestion for recoding 7 categories to 5 does not have equal distances, and arbitrarily treats answers of 1 or 2 on the 7-point item as equivalent, even though 6 and 7 are different.  There's no need to throw away the distinction you asked respondents to make between answers 1 and 2. 

You say you will be using AMOS or PLS for the analysis.  That indicates you are not bothered by treating Likert items as if they were interval measures. That's very commonly done, and only an extreme purist would object.  The good news is that the SEM software will adjust for the problem of unequal dispersion of the items as part of the modeling.  The loading for each item, in the SEM output, will assure that the items are given equal weight (unless you explicitly constrain the model to do something else). The z-score or recode fixes are only needed if you're constructing an index for other analyses, outside of the SEM framework.

I hope this is helpful.  Good luck with your project. 

Thank you so much for all the answer provided...It really helps....! 

What literature can be used to justify using the conversion table mentioned by Terans?

You can convert all item scores into ten point interval scale.

You should probably best analyze the multiple response item separately from the other 4. But if you absolutely must put them all on the same scale, you could treat each of the 11 possible responses to the multiple answer question as a separate Yes/No item, and score each one of them as 1 for Yes and 5 for No (or 1 for No and 5 for Yes if you prefer).

Yadgar, you could score the preferred responses as 1 and the non-preferred responses as 0. Then add the scores.

You add the individual item scores; you will end up with a scale score of 0 - 5. If you are looking for computer code you would first need to recode all the item variables to dichotomous variables (0=non-preferred responses, 1 = preferred responses). Then you would create a scale score variable which will be the sum of the five recoded item variables (add item 1 + item 2 + item 3 + item 4 + item 5).

Maybe I misunderstand your issue, but it sounds to me like you have four Likert items, each scored 1 to 5, along with one "check all that apply" question with 11 choices, and you want to combine them all into a single scale or index. The psychological task of "check all that apply" is quite different from the task of "rate these from Strongly Agree to Strongly Disagree." That is why I do not recommend combining all of these into a single scale. I would suggest you sum the four Likert items (assuming they are all scored in the same direction) to form a scale, and separately analyze the "check all that apply" question. If the latter is closely related, conceptually, to the Likert items, you might use the check-all responses to help validate the scale, rather than combining everything. If you absolutely must combine all of them into a single scale, treat the check-all options as 11 separate items, each one scored 1 for Yes and 5 for No (or the reverse, as appropriate). Then sum the 4 Likert items and the 11 check-all options, to form a 15-item scale. I think that would be better than collapsing the 4 Likert items to form dichotomies, each scored 1/0. I see no reason to discard the ordinal information in the Likert items just for the sake of giving them a comparable metric (same range) as the Yes/No choices on the check-all question. Scoring the check-all options as 1/5 gives them the same range as the Likert items, without discarding information in the Likerts.

Thanks for your direct message, clarifying that the 11-choice item is not check-all but instead has 11 ordered categories.

In that case, my original post in this thread, last year, provides you with guidance. If you want to combine all 5 items in a scale, you can simply compute the z-scores of all 5 items, and sum those to form the scale.

Or, if you don't like calculating z-scores for ordinal variables, you can recode the 4 Likert items and the 11-category item to have the same metric, as follows: Recode the 11 categories so they are scored zero to 10 instead of 1 to 11. Recode each of the Likert items so they are scored zero, 2.5, 5, 7.5, 10, instead of 1,2,3,4,5. Add up all 5 of the recoded items, yielding a scale with a minimum possible score of zero and a maximum possible score of 50.

Either of these two alternative approaches retains all of the original information in each of the items, maintains equal-appearing intervals between categories for the Likert items (and also for the 11-category ordinal item), and sets all of the items to have similar dispersions.

Either way, you get a single scale summarizing the results. You could use Cronbach's alpha to examine whether it makes sense to combine all of the items or not. The "standardized-item alpha" would be right for the z-score approach; plain alpha (on the recoded items) would be appropriate to the other approach.

I hope this helps. See my post from last year for further discussion.

Pallant Julllie 2010 can also be resourceful

Hi, Would anyone be able to tell me the compute function to use in SPSS to convert an already populated 4 point likert scale to a 3 point scale?

Appreciate any help.

I have currently a 9 point likert scale and a 4 point. I want do do a moderation study with these two scales but I need to convert them to the same size now.

I realise I should have done this in the beginning.

Hi, Susan Califf . Transform/Recode Into Different Variable would also work, as would the Transform/Compute mentioned by Yadgar M.Hamakhan

Thanks so much for the reply. The issue with retaining the 4 is no middle point for analysis. I want to do a correlation and possibly a Mann-Whitney test as they are ordinal data, I hope I am on the correct track.

it will not affect your results, but it is good to use 7 point than 5 point in bringing more accuracy in results.

@ Irshad, more point is always better than less so I prefer 7 point scale. It will explore the area in more depth and also gives you an opportunity to develop further unidentified/new areas of investigation. Good luck

Any suggestion to recode 1-5 in 0-3?

Thanks :)

Please note that anything other than a 5 or 7 point scale is not a Likert scale. That does not mean that it is a bad scale but it will not have the psychometric characteristics of a Likert scale. Any scale with an even number of response choices is a forced choice scale. There are times when forced choice is appropriate and times when it is not. You need to think about why you are using a particular scale. For reporting purposes, it is ok to collapse categories as long as it is clearly stated what, how and why you did it.

Call the 1-5 item you want to recode X. Call the recoded version NEWX. If you want NEWX to be scored from 0 to 3, to retain all the original information that your respondents provided in X, and to retain equal-appearing intervals between adjacent categories, you can use the following conversion equation:

NEWX = (3/4)(X - 1). That yields a recoded version scored 0, .75, 1.5, 2.25, 3.

The factor (3/4) gives the desired non-integer values with equal-appearing intervals. That's because a 1-5 range has 4 intervals between adjacent integer values, while a 0-3 range (or, equivalently, a 1-4 range) has only 3 integer intervals. Comparing the number of integer intervals to determine the conversion factor works quite generally. For example, my first posting in this thread in 2017 recoded an item scored 1 to 5 (4 integer intervals) to scores of 1 to 7 (6 integer intervals), as follows: 1, 2.5, 4, 5.5, 7. The implicit conversion equation there was

RECODEDX = 1 + (6/4)(X -1).

Burke D. Grandjean thank you

@Burke D. Grandjean In my research, I use a 4-point Likert, 5-point Likert, 6-point Likert scale. I calculated the equivalence of the scale scores with the help of the formula. But how can I do this in Spss?

Should I do as we convert inverse items in the scale using the transform menu? Transform - recode into different variable - old and new value.

Old value 1-new value1

Old value 2-new value 2.2

Old value 3-new value 3,4 ..... Ok. shaped.

Do I need to use Range?

Another question is, will this process be applied to the data collected from the participants for the variable or will the mean values of the variable be calculated and these mean values converted?

Tülin Ögül , yes, you can do the formula by hand and then do Transform, Recode (into either same or different variable, as you prefer). Or you could do it all in one step using Transform, Compute and specify the formula within the compute statement.

I'm not entirely sure I understand your second question. Either Transform, Recode or Transform, Compute applies to the data collected from the participants and will, therefore, also affect the mean values of the variable.

In most cases, you would want to apply the process to the data collected from the participants, so that the resulting transformed variable would be available for whatever other statistical procedures you might want to use for that variable. But if for some reason you only care about making comparisons of means you have already calculated without transforming the variable, you could apply the formula to the mean alone. The formula is a linear transformation; any linear transformation of scores transforms the mean in the same way.

Note that 4-point and 6-point are not Likert scales. You cannot transform them into Likert scales because you did not allow your respondents a neutral point.

Burke D. Grandjean Thank you for this explanatory information.

So is it necessary to convert your scales let's say 7 to 5 point Likert if you are using 5 point Likert scale for most of your other variables in the study?

I think my original response, May 17 2017, answers your question. In short, it is usually not necessary, often ill-advised, sometimes convenient, and occasionally important. See details in my earlier response.