how is it possible to have beta coefficient as zero while the variable is highly significant according to p-value in binary logistic regression. I am facing this problem in my project on fraud analysis. though my model is doing good on test data. it is misclassification rate is 10% only.I even tried to find out standarised beta coefficient but my model looses its predictive power on test data. there is drastic increase in misclassification rate.
Best Answer
@whuber is right, we need to know more. I will give you two guesses, but they are only guesses as we have insufficient information. Both guesses assume that your parameter is not 0 but something small, maybe so small that for display purposes your software rounded them to 0.
First guess: your parameter is a odds ratio. An odds ratio of close to 0 is a huge (negative) effect, so no surprise that that would be significant. An odds ratio is a ratio, so no effect corresponds to an odds ratio of 1.
Second guess: your parameter is a log-odds ratio, but the scale of your explanatory variable is tiny. Say Gross Domestic Product (GDP) in dollars. GDP can have a strong effect, but a single dollar increase in GDP is so tiny that that change may still lead to a change in log odds close to 0.
Similar Posts:
- Solved – how is it possible to have beta coefficient as zero while the variable is highly significant according to p-value.
- Solved – Relationship between logit and odds ratios
- Solved – How to interpret the odds ratio of an interaction term in Conditional Logistic Regression
- Solved – Interpreting mlogit coefficients in R – odds ratios and negative coefficients
- Solved – Interpreting mlogit coefficients in R – odds ratios and negative coefficients