Logistic Regression model: a significant predicting variable with low sensitivity?

VioV

New Member
#1
Hi fellow forum members! Thank you in advance for your time! I used binary logistics regression (SPSS) to determine the relationship between ambient noise levels (a continuous variable on a logarithmic scale, dB) and a single dichotomous dependent variable ("yes" - 1- for whale positive detections, and "no" - 0 - for negative detections). The model was statistically significant overall, and the Odds Ratio shows that the independent continuous variable is a significant predictor of whale detections. However, although the specificity of the model is very high (98%), the sensitivity is very low, with an extremely low percentage of true positives (8%). My question is: if detections (my dependent yes/no variable) show a significant positive association with noise level, how can sensitivity be so low? Should I trust these results? Thank you for any suggestions! I am quite new at interpreting logistic regressions.
 

rogojel

TS Contributor
#2
Re: Logistic Regression model: a significant predicting variable with low sensitivity

hi,
I assume you used some decision rule to identify a positive whale detection (like probability > 50%)? Could it be that the rule is the problem?
regards