I am analyzing the rates of malnutrition (wasting) in two datasets of screenings that were previously held. In the first dataset (with over 600 children), the wasting rate was 9.5% and the second (with over 1300 people) was 14%. I would now like to determine the probability of a malnourished child being detected in each of the screenings (the first and second). Any assistance on this will be greatly appreciated.
Dear Emmanuel Nene Odjidja ,
I think the predictive power of any model will depend upon what other data (independent variables) you are collecting about these children.
You could try a binary logistic regression on each dataset, or pool the data or use the model derived from one on the other. Your independent variables need to be binary or scale. There are assumptions they need to meet. You can measure the % of correct predictions or use the R square equivalent statistics.
ideal would be binary logistic model. The fitness of the model (R square) should good enough for prediction any R square value of 60 % will be a useful model
For a binary logistic regression, besides using the ROC curve to judge the adequacy of the model, I recommend also using the Sensitivity and Specificity (at the optimum cut-point) as well as the Misclassification Rate and the Positive and Negative Predictive Values. The predictor variables may be discrete, continuous, or categorical.
- Badges
- Science topic
- Similar topics
- Medicine
- Epi
- Age Groups
- Children