First, I have never heard of the intervals that Abbas Lafta Kneehr suggests, and they seem rather arbitrary to me.

Second, there is a difference between Likest-Scored items (which are ordinal and thus have a median rather than a mean) versus scales that are created from multiple Likert-scored items (which can be considered interval if they have sufficient reliability, and can have mean).

Hello John,

With regard to your first question, I think you simply need to provide headings at the top of your columns in a table to indicate what is entered under each of those headings. Please feel free to come back to me if you don't understand what I mean.

With regard to your second question, I think that the article I have attached should be helpful. Norman argues his points very nicely.

All the best with your research.

Dear John Peter Okoro

I expecte you need the following computations:

The number of cells of the scale to obtain the correct cell length (3/4 = 0.75) after this value was added to the lowest value in the scale (or the beginning of the scale is the correct one) to determine the upper limit of this cell, and thus became the length of cells as follows:

1 to 1.75 represents (never applicable) towards each statement, depending on the axis to be measured.

- from 1.76 to 2.50 represents (rarely applicable) towards each statement according to the axis to be measured.

2.51 to 3.25 represents (often applicable) towards each statement according to the axis to be measured.

- From 3.26 to 4.00 represents (strongly applicable) towards each statement according to the axis to be measure.

If you need unlike that you can write what you need exactily.

Good luck.

First, I have never heard of the intervals that Abbas Lafta Kneehr suggests, and they seem rather arbitrary to me.

Second, there is a difference between Likest-Scored items (which are ordinal and thus have a median rather than a mean) versus scales that are created from multiple Likert-scored items (which can be considered interval if they have sufficient reliability, and can have mean).

David L Morgan

Thank you very much for your very helpful response.

However, one of the tables; the very one with 28 items sought responses on common crimes in the study area. So the intention is to report which of the crimes are common based on the view of the respondents in the study. Now the question is, since the items are in composite form, is it acceptable to treat it as interval scare by calculating average and concluding based on cut-off points like Abbas Lafta Kneehr suggested? Thanks

No, the individual items are ordinal, and cannot be added and divided to produce a mean. I suggest you produce one table with the 28 items, and then report the mean and standard deviation for scale in the text (that information is so basic that it does not require a full table).

Thanks, Dr. David L Morgan

But, since the table might exceed one page, I hope no problem?

Thanks Dr. Abbas Lafta Kneehr

Please, do we have any literature to support that computation procedure?

I will be grateful to have it please. Thanks

Thanks Dr. Robert Trevethan

The literature is very helpful!

John Peter Okoro. You're welcome. Please feel free to get back if you have any further queries. You are probably worrying about things that others of us have worried about some time ago, solved, and happily moved ahead with the decisions that we made back earlier.

Dear John Peter Okoro

I have Only Arabic literature.

The idea is calculate the range

Max order (4) - mini order(1)=3

And then calculate the lengh of interval by dividing the range Over number of order 4(1,2,3,4) that is mean (3/4=.75) after that we can determine the uper and lower limit of each interval (1+.75=1.75)so the first interval (1 to 1.75) and the second is (from 1.76 to 2.50) and so on.

Also you can use the hypothesize mean whic can calculate as follows :

Sum the order points (1+2+3+4=10) and dividing these sumation on the number of order point(4)

10/4=2.5 the hypothesize mean so we can compare our weighted mean with the hypothesize mean (greater than or less than) also we can use this hypothesize mean to run t-test of each deffrenc between (weighted mean and hypothesize mean) by using spss software.

The following atached picture is table of testing the significance of deffrenc between weighted mean and hypothesize mean if t-test value is posative and significance then we conclud that our weighted mean is greater than hypothesized mean (2.5) significally and if the value of t-test is negative or not significance then we conclud our weighted mean is not important.

John Peter Okoro . I am sure that Abbas Lafta Kneehr is trying to be helpful, but I have to confess that I don't understand how what he is suggesting relates to your particular issues. I would also caution you against using statistics such as t-tests on individual items. In the literature, I have see that kind of activity referred to a "laundry list fuzzy jumble". It happens quite often but runs counter to the notion that scales are intended to be the composite of a number of items.

I hope that's helpful - and I'd be quite happy if others bring forward contrary advice. We all live and learn.

As others have said, the mean of a 4-category Likert item is hard to defend as a meaningful summary statistic, because such an item does not come close to providing interval-level measurement. If you want a single number to summarize the central tendency of such an item, you can dichotomize it (e.g., Strongly Agree vs. all other responses, or Strongly Agree and Agree vs. Disagree and Strongly Disagree). Then the % in one end of the dichotomy concisely summarizes the central tendency. You can readily do statistical inference on that % (confidence intervals, significance tests). As long as your sample size is reasonably large, a z-test for difference of proportions (or %s) will yield essentially identical results to a t-test for difference of means. By dichotomizing, you discard some of the detail in the 4 categories you asked your respondents to report, but the gains in simplicity of reporting (a single % per item, not 4 %s for each item) and in options for analysis (t-tests, logistic regression, etc.) will often more than outweigh that loss.

Burke D. Grandjean. Thank you for contributing to this thread. You have given me a perspective that I had abandoned but think I should have retained, namely that dichotomising data (though often disapproved of because of "loss of information") might have advantages that outweigh the disadvantages.

Everything Burke Grandjean would apply to every dichotomization of a 4-point Likert scale (e.g., 1 versus 2-4 or 1-3 versus 4), so you would need to have a good reason to cut the scale at 1-2 versus 3-4.

Exactly Prof Morgan... Prof Burke, I agree with you. There should be a neutral scale on the Likert scale to foster a good measurement.

Thank you all for your helpful contributions

Dear John Peter Okoro

The following atached file contain in page 53 the (RII) law. For five points Likert scale. The literature you need it.

Thank you very much Dr Abbas Lafta Kneehr