# Thread: Odds ratio using the negative log-log link functin in ordinal logistic regression

1. ## Odds ratio using the negative log-log link functin in ordinal logistic regression

Hello,
Below is a description of a statistical problem and questions for which I need clarification.
The data I'm working on are from an online survey (1 measuring time) with adolescents, n = 750. I work with SPSS and my boss is also functional with MPlus.

My research question is to determine the variables associated with an increased frequency of teasing/bullying experienced on different types (overweight, underweight, sexual orientation, intelligence, disability, other). The dependent variable (DV) is ordinal 1 Never, 2 Rarely, 3 Sometimes, 4 Often, 5 Very Often.

The distribution of the DV is exponentially decreasing whereby the majority never suffered ridicule and very few suffered very often. With this distribution for the DV, arises a problem of observed frequencies per cell when crossing it to the independent variables. To minimize it, I have thus grouped the DV into three categories (1) Never - 0, (2 & 3) Occasionally - 1 and (4 & 5) Frequently - 2. Is the coding of the DV should rather be 0, 1, 2 or 2, 1, 0?

At first, I did chi-square analyses between the type of teasing and sex to see if girls and boys differently suffer bullying, and compared the adjusted Z scores.

As I plan to do ordinal logistic regressions building (counting on correlational links), I excluded sex of subsequent analyses. So I have, on one side of the equation) the three levels of bullying, and the other side, academic performance in French and Physical Education (2 4-point ordinal variables dichotomized into No problems (0 ) or Problems (1)), self-concept about physical appearance and self-esteem (2 continuous variables rescaled by the interquartile range to ease the interpretation).

Besides, this interpretation is causing me problems. First, given the distribution of my DV, I have to select the negative log-log link function in SPSS PLUM options. I wonder if the formula to calculate the odds ratio is the same as for the logit link. It seems that we should go through Utilities / OMS Control Panel to get the odds ratio. Is that correct? Moreover, I wonder whether the interpretation in terms of increasing probabilities is the same. Chan (2005) said that for link functions other than logit, interpretation may not be as direct (p. 265).

Finally, there is the question of the type of comparisons carried out by the software. O'connell (2006) in her book in the Sage series, available at the following address:
http://hbanaszak.mjr.uw.edu.pl/TempT..._Sciences_.pdf
shows a table on page 30 where she compares three models: cumulative odds, continuation ratio and adjacent categories. I’m not sure what fits my needs or how to program it in SPSS. I'm interested in the characteristics of those who are bullied to see what distinguish them; so I want to compare Frequently to Occasionally, Frequently with Never, and Occasionally with Never. I’m also unsure about the influence of the choice of these above three models on the interpretation of results.

Here was my series of questions.
Thank you very much in advance for your help. I have searched a lot ( O'connell ; Field ; Tabashnick & Fidell , Internet , colleagues ) and I haven’t really found the answers to my questions yet.

Sincerely
Genevieve Lepage
Postdoctoral researcher at
Université du Québec en Outaouais

PS: I attached an examplar SPSS output.

2. ## Re: Odds ratio using the negative log-log link functin in ordinal logistic regression

Bump. This was in the moderation queue (probably because of the link) but it looks fine to me. Bumping for exposure.

 Tweet

#### Posting Permissions

• You may not post new threads
• You may not post replies
• You may not post attachments
• You may not edit your posts