Share
Explore BrainMass

Variable Targdols

This problem analyzes the asc.xls data set. The purpose of this problem is to predict which customers will respond to a particular catalog and how much they will spend. The variable targdols gives the amount that each customer spent in response to a particular mailing. The variables recency, totfreq, and totfreq give RFM for the customer at the time of the mailing.
1. Compute a new variable aos = totdols/totfreq. This variable gives the average order size. This variable is often used in place of total dollars because total dollars is usually highly correlated with frequency.
2. Compute boxplots of the variables recency, totfreq, totdols, targdols, and aos. Comment on the shape of each.
3. Compute logs of each variable. For the totdols, targdols, and aos variable you will have to add 1 before taking logs, i.e., logtarg = loge(targdols+1). This is not necessary for recency and totfreq. Also compute a variable targact=targdols>0, which takes the value 1 if a customer was "actived" and 0 if the customer did not respond.
4. Compute boxplots of the variables logrec, logfreq, logdols, logtarg, and logaos. Comment on the shape of each.
5. Compute boxplots of logtarg for each level of targact. To do this, click on Graph / Boxplot. Copy logtarg as the Y variable and targact as the x variable. This will give you separate boxplots for responders and nonresponders. Comment on the shape of each.
6. Fit a logistic regression of targact on logrec, logfreq, and logdols. Report the estimated regression equation and the log likelihood value.
7. Fit a logistic regression of targact on logrec, logfreq, and logaos. Report the estimated regression equation and the log likelihood value. Save the predicted probabilities.
8. Which model is better?
9. For responders only, use linear regression to predict logtarg from logrec, logfreq, and logdols. Report the estimated regression equation and the value of R-squared. Save the predicted values. Hint: click on Stat / Regression / Regression. Copy logtarg into the response box and logrec, logfreq, and logdols in the predictors box. Click on Options and copy the targact variable into the Weights box.
10. Compute the expected dollar amount that a customer will spend on this catalog by multiplying the probability of response and exp(fitted value + MSE/2). The fitted value comes from the linear regression and MSE comes from the ANOVA table. Report the mean of this variable in a short sentence or two.

Attachments

Solution Summary

This problem analyzes the asc.xls data set. The purpose of this problem is to predict which customers will respond to a particular catalog and how much they will spend. The variable targdols gives the amount that each customer spent in response to a particular mailing. The variables recency, totfreq, and totfreq give RFM for the customer at the time of the mailing.
1. Compute a new variable aos = totdols/totfreq. This variable gives the average order size. This variable is often used in place of total dollars because total dollars is usually highly correlated with frequency.
2. Compute boxplots of the variables recency, totfreq, totdols, targdols, and aos. Comment on the shape of each.
3. Compute logs of each variable. For the totdols, targdols, and aos variable you will have to add 1 before taking logs, i.e., logtarg = loge(targdols+1). This is not necessary for recency and totfreq. Also compute a variable targact=targdols>0, which takes the value 1 if a customer was "actived" and 0 if the customer did not respond.
4. Compute boxplots of the variables logrec, logfreq, logdols, logtarg, and logaos. Comment on the shape of each.
5. Compute boxplots of logtarg for each level of targact. To do this, click on Graph / Boxplot. Copy logtarg as the Y variable and targact as the x variable. This will give you separate boxplots for responders and nonresponders. Comment on the shape of each.
6. Fit a logistic regression of targact on logrec, logfreq, and logdols. Report the estimated regression equation and the log likelihood value.
7. Fit a logistic regression of targact on logrec, logfreq, and logaos. Report the estimated regression equation and the log likelihood value. Save the predicted probabilities.
8. Which model is better?
9. For responders only, use linear regression to predict logtarg from logrec, logfreq, and logdols. Report the estimated regression equation and the value of R-squared. Save the predicted values. Hint: click on Stat / Regression / Regression. Copy logtarg into the response box and logrec, logfreq, and logdols in the predictors box. Click on Options and copy the targact variable into the Weights box.
10. Compute the expected dollar amount that a customer will spend on this catalog by multiplying the probability of response and exp(fitted value + MSE/2). The fitted value comes from the linear regression and MSE comes from the ANOVA table. Report the mean of this variable in a short sentence or two.

$2.19