Your interpretation is close, but the correct interpretation is that the odds of sharing information decrease by 76.9% for a one-unit increase in sentiment intensity. The "one-unit" in this context refers to the change from 0 to 1 on the sentiment intensity scale. The Odds Ratio (OR) is a measure of association between an exposure and an outcome. In a logit model, it represents the factor by which the odds of the outcome (in this case, sharing information) change for a one-unit increase in the predictor variable (sentiment intensity). An OR of 1 indicates no change in the odds. An OR greater than 1 means the odds increase, and an OR less than 1 means the odds decrease.

In this case, with an OR = 0.231:

The odds of sharing information are 0.231 times the odds of not sharing information for every one-unit increase in sentiment intensity.

The percentage change in the odds is calculated as (OR - 1) x 100%.

(0.231−1)×100%=−0.769×100%=−76.9%

This means that for a one-unit increase in sentiment intensity, the odds of sharing information decrease by 76.9%.

Your interpretation incorrectly substitutes "probability" for "odds." While related, they are not the same. Odds are the ratio of the probability of an event happening to the probability of it not happening, whereas probability is the likelihood of an event occurring.

You can read "Discovering Statistics Using R" by Andy Field", for more clarification.

Regards

Since a 1-unit change is meaningless on your predictor variable (sentiment intensity has a range between 0 and 0.3, so a 1-unit change is difficult or impossible to interpret), it may be good to think more in terms of what the expected/estimated change in odds is for a 0.1-unit change (or an even smaller unit change) in the predictor variable.

Visualizing the predicted response is one of the most effective ways to understand/communicate the result in meaningful probabilities (IMHO).

You could generate a sequence of sentiment intensities from 0 to 0.3 (0.01, 0.02, ... 0.3) and use the fitted statistical model to predict the probability of sharing for each of these values. If there are relevant variables, they should be set to the median or other meaningful value(s). You can then plot "probability of sharing" vs "sentiment intensity scale" with confidence intervals (example attached).

Plotting the predicted probabilities lets you see the practical relevance of the effect. The difference in probability change per 0.1 unit sentiment intensity depends on the baseline probability of sharing and will be greatest around 0.5 and smallest around 0 or 1. Example, if the mean probability of sharing is 0.00024, even an odds ratio of 2 will most likely not be relevant in practice.

Jochen Wilhelm

Thank you very much for all the insightful perspectives you’ve provided. I have carefully read through the discussions and strongly agree with your viewpoints. However, I would like to pose a more straightforward question regarding the issue of "how to describe this result."

Can I convey this result to the readers using the following statement: "This means that for a one-unit increase in sentiment intensity, the odds of sharing information decrease by 76.9%"? Is there a more suitable way to phrase it? Perhaps, as Christian Geiser suggested, we should focus on changes of 0.1 units (where every 0.1 unit increase corresponds to a decrease in odds of 7.69%)? Jochen Wilhelm

Thank you! Jochen Wilhelm

Fan Chao, if the sentiment intensity can only vary between 0 and 0.3, I think it makes more sense to talk about a 0.1 change. Re-iterating, how much the probability changes depends on whether you go from 0 to 0.1, 0.1 to 0.2, 0.15 to 0.25, etc. Ideally, you would report it for one or more scenarios that make sense. Most of the time one would use something like the median sentiment intensity but if one is to use results to do interventions, at what sentiment intensity would you make an intervention, 0.05, 0.1 or some other "low" value?

Also to second the responses from Jochen Wilhelm

Thanks for your kindly reply! Gi-Mick Wu