0

I am struggling with the interpretation of my results, specifically the right wording to convey the output. I have looked at multiple examples on this site, but still have not found clarity. I believe my confusion is in the fact that all of my IVs are categories (gender, race, ethnicity, etc. along with the use of a specific program), so there isn't an increase by one that is possible. My DV is number of suicide deaths (so something I want to write up clearly). This is an example of the use of the program on gender with male as the reference group: 95% Confidence Interval Parameter β SE p Exp(β) LL UL (Intercept) -4.11 0.11 <0.001 0.02 0.01 0.02 Sex = Female -1.38 0.11 <0.001 0.25 0.20 0.31

I see that exp(-1.38)=0.25, but where my struggle is how to conceptualize that there is a decrease in the predicted number of suicide deaths for females using the program by 0.25. Can anyone provide clarity on how I convey the multiple levels of categories? Thank you for helping me! This type of regression is very new to me.

Debra Perez
  • 3

1 Answers1

0

Here is your output in a cleaner format.

β SE p Exp(β) LL UL
(Intercept) -4.11 0.11 <0.001 0.02 0.01 0.02
Sex = Female -1.38 0.11 <0.001 0.25 0.20 0.31

If there are other main effects in this model like race, ethnicity, and program, the coefficient for Sex=Female would be interpreted as follows. Adjusting for race, ethnicity, and program, the mean suicide rate in females is estimated to be 25% of the mean suicide rate in males. We are 95% confident the true rate ratio in the population is between 0.20 and 0.31. You may find it easier to interpret a ratio greater than 1. Adjusting for race, ethnicity, and program, the mean suicide rate in males is estimated to be 4 times higher than the mean rate in females. We are 95% confident the true rate ratio is within (3.2, 5).

I suggest you examine the modeled mean and see that it is different in each race, ethnicity, sex, and program category, but that the estimated mean for males is always 4 times larger than that for females.

If your interest is in the effect of a program on suicide then you would interpret its coefficient in the same manner described above. Adjusting for sex, race, and ethnicity, the mean suicide rate for those not taking the program is estimated to be xxx times the rate for those taking the program.

Geoffrey Johnson
  • 2,800
  • Thank you so much! And thank you for fixing the table. I copied the upper portion of my table, but it clearly did not come across the way it looked when I pasted it. That makes so much more sense. I kept getting hung up an increase in gender which is obviously not possible. The data I ran uses program implementation and gender or race or ethnicity or cause of death or location as predictors of the numbers of deaths by suicide. I did run the program on its own without any demographic factors as well. I really appreciate your time. – Debra Perez Aug 05 '21 at 23:59
  • Happy to help! On youtube everyone says, "Don't forget to like and subscribe!" Here I'll say, "don't forget to upvote and select as the best answer!" – Geoffrey Johnson Aug 06 '21 at 00:05