# How to interpret the odds ratio (exp(b)) with a continuous dependent variable?

#### FeniaWy

##### New Member
I have applied a generalized linear model with odds ratio (OR) and 95% confidence interval (CI) in order to determine the association between the total number of infarcts with cardiovascular risk factors
How I should interpret the OR with a continuous dependent variable?

#### FeniaWy

##### New Member
No with Generalized linear model. How should I interpret the Exp(B) in the output?

#### obh

##### Active Member
Hi Fenia,

Can you please write the equation you used?

#### Dason

Logistic regression is a generalized linear model which uses the bernoulli distribution as the response family and the logit as the link function.

#### hlsmith

##### Less is more. Stay pure. Stay poor.
So I get what model you are using, but did not follow the formatting of the variables. Total number of infarctions per person? So you are writing about a count variable?

#### FeniaWy

##### New Member
So I get what model you are using, but did not follow the formatting of the variables. Total number of infarctions per person? So you are writing about a count variable?
Yes exactly the dependent variable is a continuous variable (total number of infarcts per person) and the independent variable is sometimes categorical (e.g. hypertension yes vs. no: does the person has hypertension or not) and sometimes continuous as well (e.g. age of the participant)

#### FeniaWy

##### New Member
So I get what model you are using, but did not follow the formatting of the variables. Total number of infarctions per person? So you are writing about a count variable?
Yes exactly the dependent variable is a continuous variable (total number of infarcts per person) and the independent variable is sometimes categorical (e.g. hypertension yes vs. no: does the person has hypertension or not) and sometimes continuous as well (e.g. age of the participant)
Logistic regression is a generalized linear model which uses the bernoulli distribution as the response family and the logit as the link function.
Thanks a lot! Do you know how to interpret the odds ratio if the dependent variable is continuous?

#### GretaGarbo

##### Human
the dependent variable is a continuous variable (total number of infarcts per person)
But then the dependent variable is not a continuous variable. It is a discreet variable, with values like 0, 1, 2, 3, 4,.....
But maybe you mean ratio scale. So the four infarcts are twice as many as two infartcs. But for an interval variable like temperature four degrees of Celsius in not twice as much as two degrees of Celsius.

For the variable number of infarcts it would better to specify a Poisson distribution with a log link (and skip the talk about odds ratios, which just seems to confuse).

#### obh

##### Active Member
Hi Fenia,

Can you write a short example data, and what spss function do you use?
Maybe also the spss output of the example

Last edited:

#### hlsmith

##### Less is more. Stay pure. Stay poor.
I was leading you up to what @GretaGarbo posted. Though before straight jump to the Poisson Regression, can you post a histogram of your dependent variable and let us know your sample size.

Thanks!

#### FeniaWy

##### New Member
As far as I remember, a 1-unit increase in the predictor increases/decreases the odss for the positive outcome of the DV by x.
I happened to use LR (with both continous and categorical predictors) here:
https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0192039

You may want to have a peek just to see how I interpreted the ORs.
Thanks a lot, this helps already to understand it a Little bit better, but your dependent variable is categorical and mine is continuous :/ do you know how to Interpret the exp(b) if the dependent and the Independent variables are continuous?

Thanks!

#### GretaGarbo

##### Human
but your dependent variable is categorical and mine is continuous
No, it is NOT continuous.

It is a discreet variable, with values like 0, 1, 2, 3, 4,.....
(If people does not learn when we try to inform....)

#### hlsmith

##### Less is more. Stay pure. Stay poor.
@GretaGarbo is right, that counts are discrete. Meaning you can't have 3.33 infarcts. You have count data, which should be model with a count model. Which count model may be dependent on what the distribution of the counts looks like. Many times Poisson is used, but if you have many 0 values or many people with high counts other models can be used. Side note, Poisson regression will provide you with relative risks not odds ratios. Or you can have it create risk differences, which should be the standard in medicine.

What is your sample size and design of the study? Also provide a distribution of the outcome variable (histogram).

Thanks.