Dear Farimah,

Once the reference category is women (0 females) and the coefficient sign is negative, it means that men towards women have significatly less social interaction . Note: But if you want to compare your model acording do gender, you could use it as a moderator variable  (multi-group CFA)...

Have a nice time

Helena

Dear Farimah,

Once the reference category is women (0 females) and the coefficient sign is negative, it means that men towards women have significatly less social interaction . Note: But if you want to compare your model acording do gender, you could use it as a moderator variable  (multi-group CFA)...

Have a nice time

Helena

Dear Prof. Pestana

Thank you very much for your reply.

I am not comparing the model according to gender. Gender is just an independent variable. There is another dummy 2-category variable in the model as well.

Indeed, I want to know which predictors are more powerful in the model. The model is including 2 ordinal reflective IVs,   2 categorical dummy IVs, 3 continuous formative IVs, and 1 ordinal reflective DV.

For multiple regression, we were supposed to write the equation. I do not know how it should be written in this case!

For example, is it right if I say: one unit increase in the number of male respondents leads social interaction to be decreased for 0.12 units!!! Sounds strange!!!

Thank you and all the best,

one unit increase in the number of male respondents leads social interaction to be decreased for 0.12 units!!! Sounds strange!!!...Yes it is strange. But you have to think that i unit is equivalente of comparing man with female...and then you say that men towards women have significatly less social interaction.

Thank you very much dear professor. 

Best of bests,

Dear Farimanh, 

I continue not having the possibility to answer  the Rg messages...but it is always a pleasure if i could help someone like you, a very kind person.

Have a wondeful time

Helena

Thank you very much dearest professor. It is really kind of you.

Have a wonderful time too.

Best of bests,

CORRELATION COEFFICIENT: The interpretation is the same for all types of correlation coefficient. It measures the strength of the relationship between two variables in a range between: -1 < r < 1. The sign gives a direction of the relationship. From this known r, then verify if it is statistically significant.

This is an interesting case in that the dependent variable Y = ordinal. It appears you group the Y into ranked order, i.e. 1st, 2nd, etc. order or preference or something of that kind. Generally, the Y value is quantitative. Nevertheless, the appropriate correlation coefficient to use would have been ranked biserial correlation coefficient which is given by:

(1)   rrb = [2(M1 - M0)] / (n1 + n2)

where M1 = mean of the category of interest; and M1 = mean of non-category of interest.

In an ordinary case where Y = quantitative, the values for M1 and M0 are straight forward calculation for X-bar (mean of Y corresponding to 1 and 0 categories). However, if your Y = ordinal, this requires you to obtain the mean M1 and M0 under discrete context because for ordinal, you have to select (one at a time) your category of interest 1st group, 2nd group, etc. .... if related to sex, then answer how does sex explain 1st group, 2nd group, etc. and then proceed to test statistics.

CALCULATING THE MEAN: Since your Y = ordinal, the means M1 and M0 would have to follow discrete probability approach. Using Laplace's rule of succession, p = (s + 1) / (n + 2) and non-success q = 1 -p. The mean is determined then should be M1 = np. In similar manner determine M0 and then find rrb.

TEST STATISTICS: Conventionally, the test statistics may be a T-test for Y = quantitative. In this case, if M1 and M0 are obtained through discrete (binomial distribution (non-continuous probability)) a T-test would work under an assumption that you data distribution converges to CLT under DeMoivre-Laplace equation:

(2)   lim (X - np) / sqrt(npq) > Z

where n = sample size and the limits goes from n to infinity. Under the Law of Large Number this will allow us to assume normal distribution where T = Z (approximately). Test this convergence separately if your Y = ordinal and you are using T-test for r.

DATA: Double check how your Y was coded. If it is ordinal, i.e. 1st, 2nd, 3rd, etc., these are discrete (yes | no) and when cross with X = sex, this a case of ordinal Y crossing with nominal X. if the ranking was coded in Likert-type answer choice, it may have been treated as quantitative by the software.

CHECKING: If your reported T-score is 3.90 from the correlations T-test, it means that you are working with a sample of size n = 1038. is this correct? if your sample size is not this number, something went wrong with the software or data coding.

(3)   T = r(sqrt(n - 2)) / sqrt(1 - r2)

where r = -0.12 and T = 3.90, solve for n. Solution n = 1038. This is one way of checking.

Dear Dr. Louangrath

Thank you very much for your answer. 

Actually, the sample size is 412! I checked the process of independent t-test once again, nothing wrong was found!

Your reply sounds to be very complete but it is somehow complicated for me! Would you please make it clearer?

Best regards,

Dear Farimah,

Could you show a sample of your Y Question and its Response choice?

From this we could figure out easily.

Cheers.

Dear Dr. Louangrath

It is a self-developed scale measuring the level of social interaction among the residents of high-rise buildings. An example is "I spend my leisure time with familiar residents in the outdoor places". The items are on 5-point Likert scale.

Regards,

5-points scale: (1,2,3,4,5) is treated as QUANTITATIVE, not ordinal. It is called success integer with equidistance of 1. If Y = Q and X = nominal (sex), the correct correlation to use is point biserial.

Altough there are debates in the literature as how to classify the Likert scale (Q or O), ignore it. Software generally treat it as Q.

Thank you dear Dr. Louagrath.

All the best

Dear Louagrath et all, 

In terms of :

1) "Software generally treat it as Q." i would answer: software is just a program and the analyst should be the one that knows what should be done.

2) "Altough there are debates in the literature as how to classify the Likert scale (Q or O), ignore it." and i would ask you this: When do you aplly categorical methods ? Do you think they are not needed? The distance between categories in all ordinal scale should be alwasy "= 1"?  In my opinion, we should only aplly Statistical methods, after we understand what we are studying, because statisctics is only a tool to make simpler the  reality and to represent it as effectively as possible..

 Is there any 100% truth? On average there is no absolute knowledge and that is why we continue to investigate.

Have a nice time

Helena