categorical variable

[Ibraheem Khalid]

Dear researchers,
I am new to this great website, and this is my first post, and I hope to receive an answer to my questions, please.

I have a panel dataset for a set of firms for the period extending from 2000-2010. I have a categorical variable, which is a firm score, which is the dependent variable. This score consists of five scales. For instance, score 1 means very poor, 2 poor, 3 good, 4 very good, and 5 Excellent. Let’s take company A as an example. It might have a score of 3 in the year 2000, and 4 in 2001, and 5 in 2008. I mean it is not necessary to have the same code in each year for the firm.

My independent variables include continuous, binary, and categorical variables.

To be honest, I have read about which is the correct way to do the analysis, ordered logit or multinomial logit? I have found that we need to use the ordered logit regression as the dependent variable takes a number of finite and discrete values that contain ordinal information. However, the multinomial don’t contain ordinal information.

HOWEVER, the literature has mentioned that this score is changing over time and thus they can’t be compared across years. Thus, to address this issue, they have used the following equation which is:
(firm rank ‒ 1)/(number of firms e 1)
This gives the annual percentile rank of the dependent variable ranging between 0 and 1 each year. Thus, in that way, I should use the logistic regression.

My questions, please:
What is the code for performing the ordinal logit regression?
What is the code for performing the logit regression?
What is the correlation analysis code in both above two analyses?
Can I use the lag independent variables in the ordinal logit regression and the logit regression?

Many thanks in advance.

[Ibraheem Khalid]

Dear researchers,

Following the above post, I need to amend it. So, please discard the above post as I can't delete it, and please consider the below post:

I have a panel dataset for a set of firms for the period extending from 2000-2010. I have a categorical variable, which is a firm score, which is the dependent variable. This score consists of five scales. For instance, score 1 means very poor, 2 poor, 3 good, 4 very good, and 5
Excellent. Let’s take company A as an example. It might have a score of 3 in the year 2000, and 4 in 2001, and 5 in 2008. I mean it is not necessary to have the same code in each year for the firm.

My independent variables include continuous, binary, and categorical variables.

To be honest, I have read about which is the correct way to do the analysis, ordered logit or multinomial logit? I have found that we need to use the ordered logit regression as the dependent variable takes a number of finite and discrete values that contain ordinal information. However, the multinomial don’t contain ordinal information.

HOWEVER, the literature has mentioned that this score is changing over time and thus they can’t be compared across years. Thus, to address this issue, they have used the following equation which is:

(firm rank ‒ 1)/(number of firms -1)

Thus, the firm score which is the dependent variable ranges from zero for the lowest-ranked firm to one for the highest-ranked firm. That means I have converted the categorical variable to a continuous variable, and on that way, I can use multiple regression.

But what if i want to use the ordered logit, i mean before converting the dependent variable, then:

What is the code for performing the ordinal logit regression?

Many thanks in advance.

[Clyde Schechter]

Since you have panel data, you probably want to use the -xtologit- command. Do see -help xtologit- for the details.

[Ibraheem Khalid]

Thank you very much for the answer Clyde.

Actually, it seems that -xtologit- doesn't accept either fixed effect nor random effect, even does not accept the cluster error. I mean, I have tried the following:

 xtologit Y X1 X2  i.year, fe cluster (FIRM)
xtologit Y X1 X2  i.year, re cluster (FIRM)
xtologit Y X1 X2  i.year, fe 
xtologit Y X1 X2  i.year, re
xtologit Y X1 X2  i.year, cluster (FIRM)

None of the above codes worked with me

The only code that works with me is:

 xtologit Y X1 X2  i.year

I don't know if this is correct or not? What do you think Clyde?

Thanks in advance.

[Clyde Schechter]

Yes. The -xtologit- command only fits random effects models. The help file for -xtologit- makes that clear.

[Ibraheem Khalid]

Okay, now I see.
Thanks very much, Clyde for your answer.