+
Related Flashcards
Cards In This Set
| Front | Back |
|
Sensitivity
|
# of correctly predicted buyers/ actual buyers
|
|
Specificity
|
# of correctly predicted non buyers/ # of actual non-buyers
|
|
False Positives
|
# of incorrectly predicted buyers/ # predicted buyers
|
|
False Negatives
|
# of incorrectly predicted non-buyers/ # of predicted Non-buyers
|
|
The Normality Assumption is tested by
|
The plot of the cumulative distribution of the residuals vs. the cumulative standardized normal distribution
|
|
When it comes to choice, how come we can't use a linear regression model?
|
1. Since the choice variable has only two values, the error term cannot be normally distributed. 2. It would be preferable if the model predictions represent the probability that choice= yes, but this is not so in linear regression.
|
|
Odds ratio:
|
P/(1-p)
|
|
Marginal Effect
|
B*p(1-p)
|
|
Percent Concordant
|
If the predicted purchase probability for the non-buyer is lower than the predicted prchase probability for the buyer
|
|
Percent Discordant
|
The observations are discordant if the predicted purchase probability for the non-buyer is higher than the predicted purchase probability for the buyer
|
|
Models with what type of concordant are preferred?
|
Models with high concordant and low discordant are preffered over models with low concordant and high discordant
|
|
What is the characterization of the prediction from a logit model?
|
It predicted predicted purchase probability
|
|
What best describes an ROC Curve?
|
A graph of a model's sensitivity vs. 1-specificity
|
|
What is the purpose of the logit model
|
To examine associative relationships between a dependent variable and one or more independent variables
|
|
What is a logit model
|
A special type of regression in which the dependent variable is limited to a small set of possible values
|
