HELP! confidence intervals, OR.. im lost

#1
I need serious help.. I CANNOT wrap my head around statistics or what they mean/how they connect.

I have an OR of 0.88, a 95% CI of 0.72-1.08

what does all this mean in a sentence? I have been told its insignificant but why? I don't get it and neither of my parents know this stuff
 

Karabiner

TS Contributor
#2
If there is no association between variables in the population, then the odds ratio in the population would be 1.00.
In your sample, you have an odds ratio 0.88, which of deviates from 1.00. A 95% confidence interval was constructed.
Without going into details, if the 95% confidence interval includes the value expected under the null hypothesis (here:
the value 1.00 for "no association") then you cannot reject the null hypothesis ("not significant"). In the current example,
1.00 is included in the interval.

With kind regards

Karabiner
 
#5
"Statistics course and homework discussion. Elementary statistics"
Maybe close enough.
Well, I taught Statistics 101 off and on for 40 years, and am finishing up a stats book, and I can't answer or understand the question. I guess that "elementary" has changed meaning.
 

Karabiner

TS Contributor
#6
I was referring to "homework".

I'm afraid the term Statistics 101 has no further meaning for me.

With kind regards

Karabiner
 

Dason

Ambassador to the humans
#7
Honestly the subforum doesn't really matter too much. If it bothers you report it and I can move it but I don't think most people care too much
 

noetsi

Fortran must die
#8
I think the key here is an ODD RATIO of 1 is the equivalent in linear regression of an effect size of 0. So if the confidence interval contains 1 as this one did you can not be sure there was an effect at all.

Think of it this way for linear regression. If the change in Y for a one unit change in X is zero what does that mean? It has no observed impact. That is exactly what an OR of 1 means in logistic regression.
 

hlsmith

Less is more. Stay pure. Stay poor.
#9
You could ask my parents but they wouldn't know either, along with everyone in my family.

I will chime in with a few of the basics. Many times with estimates you are comparing to a null value as @Karabiner mentioned. You have a ratio of two numbers. So the traditional null (no association) value is '1'. Because if you said what are the odds of men versus women passing an exam and both had the same odds, you would be dividing the number by it's self (e.g., 3/3 = 1), which would mean no difference. Now, you have just a sample of all the people in the population you care about. So there is some doubt in the estimate you come up with since you don't have data on everyone. This is where the confidence interval comes into play. It represents if you repeatedly sampled from the population. I won't get into the intricate definitions here, but it slightly/kind of represents a range of the value (estimate for the association) that the estimate could be. So if there is a big population and I sample only a few, my estimate is likely off for the population and it could be a little higher or lower. So given the variability (sampling variability) we can't be sure the true value may not be equal to the null ('1', same value on top and bottom).
 

hlsmith

Less is more. Stay pure. Stay poor.
#10
@joeb33050 - Thanks for the input. As @Dason mentioned - I don't think anyone is gonna lose sleep over a posting in a category slightly off. This is likely better than the OPs who posts the same question in every category. I wouldn't call this elementary statistic, but it is elementary biostatistics. Thanks for your contributions - with over 40 years of stats you have likely seen quite a few advances over the years (e.g., bootstrapping, pushes toward non-parametric approaches, and software changes).
 
#11
Hello all,

I am in Engineering field, and I am new to this forum. I have a question about interpretation of confidence interval in regression analysis. It has been for a couple of days that I am trying to find the answer of my question online, but I could not find any useful information. Most of the interpretations are about samples and ... In my case, I have done some simulations and I have found an empirical equations by which I have predicted my data. Then I have plotted the predicted data versus observed data. The plot is very similar to the attached figure. The data which were used used in regression analysis are shown with black dots. The green line is the regressions line; blue lines show the confidence interval and the red ones show the prediction interval (both are 95%). My question is about interpretation of these intervals. What can I say about the data which are within the range of confidence interval and what about the ones which are between confidence and prediction intervals and the data which is out of both ranges. Any help will be greatly appreciated. Thank you. Steve.
qCn8o.png
 

hlsmith

Less is more. Stay pure. Stay poor.
#12
For clarification, Is this really a plot of observed (DV) versus Predicted (DV)? Why is one dot red, is it a predicted value?
 
#15
Should we also ignore the apparent dependence in the data?
Yes please. Assume that there is such a figure after plotting the predicted values versus observed values. All I need is the interpretation of confidence and prediction intervals in such case. What I have just found out is that, the confidence interval interpretation is not be related to the data. It is actually a range where the regression line (green line) can be. So for a given observed data, the predicted data is not exactly what is shown by the green line, but it is a range which is limited between the two blue lines. The tighter the boundary, the better the regression analysis. But the prediction interval shows the range that future observed data could fall in. For example if somebody does the same simulations, exactly with the same conditions that my simulations were conducted, he should expect to get a value with in the range of prediction interval. I am not sure how much my understanding is correct, though!