Odds ratio and p-value

jacob

New Member
#1
could anyone plz help me out....
PRESCENCE OF INURY TO OCULAR ADNEXAE
YES NO
MILD 70 12
MODERATE 13 1
SEVERE 4 2

i want to find the odds ratio and p value of each row,namely mild injury,moderate injury and severe injury....how do i do it...?
 

hlsmith

Omega Contributor
#2
Not sure you know what you really want.


First, which row will be your reference group? Since typically this table would be used to only generate two odds ratios.
 

noetsi

Fortran must die
#3
You have to actually run the data in logistic regression. No one calculates odds ratios or p values manually. You have way to little data to get meaningful responses...
 

hlsmith

Omega Contributor
#4
All right just for fun, I ran this in a logistic regression model with a penalization (firth) for small cell values, I also corrected the 95% confidence intervals using Bonferroni Correction to address false discovery. You don't need a p-value if you have confidence intervals.


Odds Ratio Estimates and Profile-Likelihood Confidence Intervals


Moderate vs Mild: 1.596 (95% CI: 0.276, 24.059)

Severe vs Mild: 0.319 (95% CI: 0.049, 2.710)


No linear trend in outcome when looking mild to severe exposure levels.
 

hlsmith

Omega Contributor
#5
OK, in all seriousness - if you just need to generate odds ratios, the first step would be to make two 2x2 tables. Ideally with the same reference group in each (bottom row). Then alphabetize the cells row 1 (A, B) and row 2 (C, D).


Next do the following (A times D) / (B times C), that is the odds ratio for that table. You will also want to add 95% confidence intervals to it. Once you do the above post your answers and we can talk about the CIs.
 

noetsi

Fortran must die
#6
And no one outside a university course will every do that :p

Its a good idea to use firth never thought of that. I did not know about the bonferoni correction.

The CI will tell you if the results are significant. They won't tell you what the specific p value is.
 
#7
And no one outside a university course will every do that :p
And what do you know about that? There are lots of companies that are doing bad things, but also many great serious companies.

I did not know about the bonferoni correction.
Yes,you did! We have talked about it many times!
(But he was a person, so should he not be spelled with an upper case B in English?)

The CI will tell you if the results are significant. They won't tell you what the specific p value is.
No, but the "CI" will tell you the confidence interval, and that is considered more informative.
 

noetsi

Fortran must die
#9
GretaGarbo I knew about the bonferoni correction. I did not know it was used as hlsmith used it.

When I said no one used this approach I meant no one manually calculated odds ratios and p values. They normally use a computer. Only in classes do you do things like manually calculate these.

Its interesting that the CI is considered more informative. In the journals I have read in the social sciences it is the p value that was stressed, the CI rarely gets mentioned. But my real point is the OP wanted to know a p value not a CI. Calculating a P value by hand would be really really hard.

Bonferoni probably should be capitalized. But I don't think it usually is much like firth is not.
 
#10
Apparently without eyebrows?
Yes, from now on it will be called the-no-eyebrows-correction.

But what about Firth? He was a person too. Don't you have love for him?
Amusingly, I can say: I did not know about the Firth correction! So, thank you, one gains and learns.



GretaGarbo I knew about the bonferoni correction. I did not know it was used as hlsmith used it.
No, you literally said: "I did not know about the bonferoni correction."


When I said no one used this approach I meant no one manually calculated odds ratios and p values. They normally use a computer. Only in classes do you do things like manually calculate these.
OK, I misunderstood you. But I would not be surprised if someone got the raw data in excel and picked out the old textbook and did the calculations "manually" with a pocket calculator of with excel.

Its interesting that the CI is considered more informative. In the journals I have read in the social sciences it is the p value that was stressed, the CI rarely gets mentioned. But my real point is the OP wanted to know a p value not a CI. Calculating a P value by hand would be really really hard.
I have heard a epidemiology professor say that "now days" (since the last 15 years, I believe) confidence intervals (CI) are more popular.
Of course a CI can be calculated from a series of tests - all the values outside of the CI can be rejected with the test, and all values within the CI can not be rejected. So in that sense CI:s and test contain the same information (there is a correspondence theorem). But people does not misunderstand the CI:s as much as p-values.

And you are right the original poster asked for p-values.
 

hlsmith

Omega Contributor
#11
The other issues with p-values is they do not show the magnitude nor the direction of association. I can say in medicine the analysts have made a huge push to get away from p-value fetishes.