In my dataset, both response and the independent variables are ordinal with multiple categories as severe (coded as 3), moderate (coded as 2), mild (coded as 1) and none (0)
I run the model by putting both outcome and the variable in descending order to see if there was a relationship between the response and the ordinal variable, and I've had the results shown below
threshold: response (3)
response (2)
response (1)
variable (1): exp (B)=3, P=0.039, 95% CI=1.061-9.253)
and the reference category is variable (0)
Could you help me to interpret the results?
It looks to me as if you are not treating the predictor as ordinal but as continuous.
It looks to me as if you are not treating the predictor as ordinal but as continuous.
I think I did not put the required info to interpret the results.
threshold (outcome=3)
(outcome=2)
(outcome=1)
predictor=3 exp (B)= 0.325, p=0.297
predictor=2 exp (B)= 0.191, P=0.009
predictor=1 exp (B)= 0.596, p=0.268
Predictor=0 B=0
(reference category is 0)
Thank you in advance!
I think most software will not handle ordinal independent variables correctly. I would check the documentation of your software, and make up some data to see if your software is actually treating the ordinal independent variable as ordinal.
If you have one ordinal dependent variable and one ordinal independent variable, you can use the linear-by-linear test. The test is described by Agresti, and is available in the coin package in R. The coin package is the only software I've encountered that handles ordinal independent variables as ordinal.
The exp (b) is calculating the relative odds to the base catgory of the predictor ; I would be very careful about this
it is preferable to convert the logits to probabilities rather than odds as you can get very counter intuitive results otherwise see 6.1.2. Interpretation of Coefficients in
STATISTICAL MODELS FOR CAUSAL ANALYSIS by ROBERT D. RETHERFORD and MINJA KIM CHOE
Given the p values you have got for predictor 3 and predictor 1, I would say that there is limited data for these two categories. You may want to re-group the categories into fewer categories
http://onlinelibrary.wiley.com/doi/10.1002/9781118033135.ch6/summary
Salvatore S. Mangiafico, Sir, cal you please elaborate a little more about the usage of the coin package in R?
As you mentioned about the using ordinal variable as ordinal is an issue, and coin package can handle it.
I have data where one or more independent variables are ordinal in nature, but when I am running OLS using them, results are coming as linear, quadratic and cubic coefficients.
Can you please help?
- Badges
- Science topic