SPSSX Discussion

gain chart

Classic

List

Threaded

10 messages Options

E. Bernardo

gain chart

I've read books on predictive models particularly predicting whether a customer churned or not. I saw a gain chart where the horizontal is percentile while the vertical is % gain, but I did not understand. I need your help for the two questions:

1. Gain was defined in the book as the proportion of total HITS that occurs in each quantile. What do you mean by "HITS"?

2. What is the interpretation of the point (40,85) on the gain chart? That is, 40% on the horizontal axis and 85% on the vertical axis?

Eins

Adding more friends is quick and easy.
Import them over to Yahoo! Mail today!

ViAnn Beadle

Re: gain chart

Hits is the number or percent of the target action occurring. In this case it was probably is churn. The gains chart plots cumulative gain by decile. So the top 40% of customers accounted for 85% of the customers churning. The more the curve varies from a diagonal line, the better the model is in identifying customers likely to churn.

This is a good example of what predictive analytics are.

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Eins Bernardo
Sent: Wednesday, August 05, 2009 4:33 AM
To: [hidden email]
Subject: gain chart

1. Gain was defined in the book as the proportion of total HITS that occurs in each quantile. What do you mean by "HITS"?

2. What is the interpretation of the point (40,85) on the gain chart? That is, 40% on the horizontal axis and 85% on the vertical axis?

Eins

Adding more friends is quick and easy.
Import them over to Yahoo! Mail today!

E. Bernardo

Re: gain chart

Thank you ViAnn for your comment. Follow-up question: How did the customers classified as being in the top 40%?

--- On Wed, 8/5/09, ViAnn Beadle <[hidden email]> wrote:

From: ViAnn Beadle <[hidden email]>
Subject: RE: gain chart
To: "'Eins Bernardo'" <[hidden email]..ph>, [hidden email]
Date: Wednesday, 5 August, 2009, 1:09 PM

Hits is the number or percent of the target action occurring. In this case it was probably is churn. The gains chart plots cumulative gain by decile. So the top 40% of customers accounted for 85% of the customers churning. The more the curve varies from a diagonal line, the better the model is in identifying customers likely to churn.

This is a good example of what predictive analytics are.

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Eins Bernardo
Sent: Wednesday, August 05, 2009 4:33 AM
To: [hidden email]
Subject: gain chart

I've read books on predictive models particularly predicting whether a customer churned or not. I saw a gain chart where the horizontal is percentile while the vertical is % gain, but I did not understand. I need your help for the two questions:

1. Gain was defined in the book as the proportion of total HITS that occurs in each quantile. What do you mean by "HITS"?

2. What is the interpretation of the point (40,85) on the gain chart? That is, 40% on the horizontal axis and 85% on the vertical axis?

Eins

Adding more friends is quick and easy.
Import them over to Yahoo! Mail today!

Importing contacts has never been easier.
Bring your friends over to Yahoo! Mail today!

ViAnn Beadle

Re: gain chart

A model was built using some type of statistical method. Usually this is done via CART or CHAID. In SPSS, you would use the TREES command.

From: Eins Bernardo [mailto:[hidden email]]
Sent: Wednesday, August 05, 2009 8:30 AM
To: [hidden email]; ViAnn Beadle
Subject: RE: gain chart

Thank you ViAnn for your comment. Follow-up question: How did the customers classified as being in the top 40%?

--- On Wed, 8/5/09, ViAnn Beadle <[hidden email]> wrote:

From: ViAnn Beadle <[hidden email]>
Subject: RE: gain chart
To: "'Eins Bernardo'" <[hidden email]..ph>, [hidden email]
Date: Wednesday, 5 August, 2009, 1:09 PM

This is a good example of what predictive analytics are.

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Eins Bernardo
Sent: Wednesday, August 05, 2009 4:33 AM
To: [hidden email]
Subject: gain chart

1. Gain was defined in the book as the proportion of total HITS that occurs in each quantile. What do you mean by "HITS"?

2. What is the interpretation of the point (40,85) on the gain chart? That is, 40% on the horizontal axis and 85% on the vertical axis?

Eins

Adding more friends is quick and easy.
Import them over to Yahoo! Mail today!

Importing contacts has never been easier.
Bring your friends over to Yahoo! Mail today!

E. Bernardo

Re: gain chart

ViAnn wrote:

A model was built using some type of statistical method. Usually this is done via CART or CHAID. In SPSS, you would use the TREES command.

The horizontal line in the chart is quantile. It simply sorts the "VALUES" from highest to lowest, then put some points P1, P2,P3, P4 ....P99 such that the top 40% of the customers fall above P4. Yes I am aware of the classification TREEs. What I didnt understand are the "VALUES"? What are these VALUES? Are these probabilities? Probabilities of what?

I am sorry for the question?

Eins

--- On Wed, 8/5/09, ViAnn Beadle <[hidden email]> wrote:

From: ViAnn Beadle <[hidden email]>
Subject: Re: gain chart
To: [hidden email]
Date: Wednesday, 5 August, 2009, 2:40 PM

A model was built using some type of statistical method. Usually this is done via CART or CHAID. In SPSS, you would use the TREES command.

From: Eins Bernardo [mailto:[hidden email]]
Sent: Wednesday, August 05, 2009 8:30 AM
To: [hidden email]; ViAnn Beadle
Subject: RE: gain chart

Thank you ViAnn for your comment. Follow-up question: How did the customers classified as being in the top 40%?

--- On Wed, 8/5/09, ViAnn Beadle <[hidden email]> wrote:

From: ViAnn Beadle <[hidden email]>
Subject: RE: gain chart
To: "'Eins Bernardo'" <[hidden email]..ph>, [hidden email]
Date: Wednesday, 5 August, 2009, 1:09 PM

Hits is the number or percent of the target action occurring. In this case it was probably is churn. The gains chart plots cumulative gain by decile. So the top 40% of customers accounted for 85% of the customers churning. The more the curve varies from a diagonal line, the better the model is in identifying customers likely to churn.

This is a good example of what predictive analytics are.

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Eins Bernardo
Sent: Wednesday, August 05, 2009 4:33 AM
To: [hidden email]
Subject: gain chart

I've read books on predictive models particularly predicting whether a customer churned or not. I saw a gain chart where the horizontal is percentile while the vertical is % gain, but I did not understand. I need your help for the two questions:

1. Gain was defined in the book as the proportion of total HITS that occurs in each quantile. What do you mean by "HITS"?

2. What is the interpretation of the point (40,85) on the gain chart? That is, 40% on the horizontal axis and 85% on the vertical axis?

Eins

Adding more friends is quick and easy.
Import them over to Yahoo! Mail today!

Importing contacts has never been easier.
Bring your friends over to Yahoo! Mail today!

Surf faster.
Internet Explorer 8 optmized for Yahoo! auto launches 2 of your favorite pages everytime you open your browser.Get IE8 here! (It's free)

ViAnn Beadle

Re: gain chart

You’re going to have to give us a reference for your question. What text are you looking at? You are now talking about “VALUES” and points P1, P2, P3—you didn’t mention any of this in your first question.

From: Eins Bernardo [mailto:[hidden email]]
Sent: Wednesday, August 05, 2009 6:42 PM
To: [hidden email]; ViAnn Beadle
Subject: Re: gain chart

ViAnn wrote:

A model was built using some type of statistical method. Usually this is done via CART or CHAID. In SPSS, you would use the TREES command.

I am sorry for the question?

Eins

--- On Wed, 8/5/09, ViAnn Beadle <[hidden email]> wrote:

From: ViAnn Beadle <[hidden email]>
Subject: Re: gain chart
To: [hidden email]
Date: Wednesday, 5 August, 2009, 2:40 PM

A model was built using some type of statistical method. Usually this is done via CART or CHAID. In SPSS, you would use the TREES command.

From: Eins Bernardo [mailto:[hidden email]]
Sent: Wednesday, August 05, 2009 8:30 AM
To: [hidden email]; ViAnn Beadle
Subject: RE: gain chart

Thank you ViAnn for your comment. Follow-up question: How did the customers classified as being in the top 40%?

--- On Wed, 8/5/09, ViAnn Beadle <[hidden email]> wrote:

From: ViAnn Beadle <[hidden email]>
Subject: RE: gain chart
To: "'Eins Bernardo'" <[hidden email]..ph>, [hidden email]
Date: Wednesday, 5 August, 2009, 1:09 PM

This is a good example of what predictive analytics are.

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Eins Bernardo
Sent: Wednesday, August 05, 2009 4:33 AM
To: [hidden email]
Subject: gain chart

1. Gain was defined in the book as the proportion of total HITS that occurs in each quantile. What do you mean by "HITS"?

2. What is the interpretation of the point (40,85) on the gain chart? That is, 40% on the horizontal axis and 85% on the vertical axis?

Eins

Adding more friends is quick and easy.
Import them over to Yahoo! Mail today!

Importing contacts has never been easier.
Bring your friends over to Yahoo! Mail today!

Surf faster.
Internet Explorer 8 optmized for Yahoo! auto launches 2 of your favorite pages everytime you open your browser.Get IE8 here! (It's free)

E. Bernardo

Re: gain chart

Sorry for the unclear explanation.

I want to understand the horizontal axis of the gain chart. The horizontal axis is "percentile", right?. How did the percentile computed? To identify the top 40% customers, is it not that the customers are sorted from highest to lowest, then identify the top 40%? What to sort? In this case I am asking what to sort? I think there is one VARIABLE used in order to sort the customers, right?. What is this variable? Sorry, I was using VALUES in my previous email instead of VARIABLE.

To further clarify, suppose the we sort the 100 customers by "amount purchased". The VARIABLE being the basis of sorting the customers is "amount purchased", right?. Once the cusromer is already sorted, we can identify the top 1%, 2%, ,,,40%,,,,99% customers (the percentiles). Clearly, the percentiles were obtained from the "amount purchased" variable. Now the horizontal axis of the gain chart is percentile. I am asking of the variable as basis of getting the percentile.

Eins

--- On Thu, 8/6/09, ViAnn Beadle <[hidden email]> wrote:

From: ViAnn Beadle <[hidden email]>
Subject: Re: gain chart
To: [hidden email]
Date: Thursday, 6 August, 2009, 1:16 AM

You’re going to have to give us a reference for your question. What text are you looking at? You are now talking about “VALUES” and points P1, P2, P3—you didn’t mention any of this in your first question.

From: Eins Bernardo [mailto:[hidden email]]
Sent: Wednesday, August 05, 2009 6:42 PM
To: [hidden email]; ViAnn Beadle
Subject: Re: gain chart

ViAnn wrote:

A model was built using some type of statistical method. Usually this is done via CART or CHAID. In SPSS, you would use the TREES command.

The horizontal line in the chart is quantile. It simply sorts the "VALUES" from highest to lowest, then put some points P1, P2,P3, P4 ....P99 such that the top 40% of the customers fall above P4. Yes I am aware of the classification TREEs. What I didnt understand are the "VALUES"? What are these VALUES? Are these probabilities? Probabilities of what?

I am sorry for the question?

Eins

--- On Wed, 8/5/09, ViAnn Beadle <[hidden email]> wrote:

From: ViAnn Beadle <[hidden email]>
Subject: Re: gain chart
To: [hidden email]
Date: Wednesday, 5 August, 2009, 2:40 PM

A model was built using some type of statistical method. Usually this is done via CART or CHAID. In SPSS, you would use the TREES command.

From: Eins Bernardo [mailto:[hidden email]]
Sent: Wednesday, August 05, 2009 8:30 AM
To: [hidden email]; ViAnn Beadle
Subject: RE: gain chart

Thank you ViAnn for your comment. Follow-up question: How did the customers classified as being in the top 40%?

--- On Wed, 8/5/09, ViAnn Beadle <[hidden email]> wrote:

From: ViAnn Beadle <[hidden email]>
Subject: RE: gain chart
To: "'Eins Bernardo'" <[hidden email]..ph>, [hidden email]
Date: Wednesday, 5 August, 2009, 1:09 PM

Hits is the number or percent of the target action occurring. In this case it was probably is churn. The gains chart plots cumulative gain by decile. So the top 40% of customers accounted for 85% of the customers churning. The more the curve varies from a diagonal line, the better the model is in identifying customers likely to churn.

This is a good example of what predictive analytics are.

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Eins Bernardo
Sent: Wednesday, August 05, 2009 4:33 AM
To: [hidden email]
Subject: gain chart

I've read books on predictive models particularly predicting whether a customer churned or not. I saw a gain chart where the horizontal is percentile while the vertical is % gain, but I did not understand. I need your help for the two questions:

1. Gain was defined in the book as the proportion of total HITS that occurs in each quantile. What do you mean by "HITS"?

2. What is the interpretation of the point (40,85) on the gain chart? That is, 40% on the horizontal axis and 85% on the vertical axis?

Eins

Adding more friends is quick and easy.
Import them over to Yahoo! Mail today!

Importing contacts has never been easier.
Bring your friends over to Yahoo! Mail today!

Surf faster.
Internet Explorer 8 optmized for Yahoo! auto launches 2 of your favorite pages everytime you open your browser.Get IE8 here! (It's free)

Bring your friends to the fun.
Invite your friends from Hotmail, Gmail to Yahoo! Mail today!

ViAnn Beadle

Re: gain chart

The variable is the propensity score computed for the target variable in the model. That propensity score identifies the best customers. Amount purchased might be such a target variable but the target variable usually is something like propensity to buy anything at all. This is not the place to discuss all this and I recommend that you read up on the literature.

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Eins Bernardo
Sent: Friday, August 07, 2009 6:39 PM
To: [hidden email]
Subject: Re: gain chart

Sorry for the unclear explanation.

Eins

--- On Thu, 8/6/09, ViAnn Beadle <[hidden email]> wrote:

From: ViAnn Beadle <[hidden email]>
Subject: Re: gain chart
To: [hidden email]
Date: Thursday, 6 August, 2009, 1:16 AM

You’re going to have to give us a reference for your question. What text are you looking at? You are now talking about “VALUES” and points P1, P2, P3—you didn’t mention any of this in your first question.

From: Eins Bernardo [mailto:[hidden email]]
Sent: Wednesday, August 05, 2009 6:42 PM
To: [hidden email]; ViAnn Beadle
Subject: Re: gain chart

ViAnn wrote:

A model was built using some type of statistical method. Usually this is done via CART or CHAID. In SPSS, you would use the TREES command.

I am sorry for the question?

Eins

--- On Wed, 8/5/09, ViAnn Beadle <[hidden email]> wrote:

From: ViAnn Beadle <[hidden email]>
Subject: Re: gain chart
To: [hidden email]
Date: Wednesday, 5 August, 2009, 2:40 PM

A model was built using some type of statistical method. Usually this is done via CART or CHAID. In SPSS, you would use the TREES command.

From: Eins Bernardo [mailto:[hidden email]]
Sent: Wednesday, August 05, 2009 8:30 AM
To: [hidden email]; ViAnn Beadle
Subject: RE: gain chart

Thank you ViAnn for your comment. Follow-up question: How did the customers classified as being in the top 40%?

--- On Wed, 8/5/09, ViAnn Beadle <[hidden email]> wrote:

From: ViAnn Beadle <[hidden email]>
Subject: RE: gain chart
To: "'Eins Bernardo'" <[hidden email]..ph>, [hidden email]
Date: Wednesday, 5 August, 2009, 1:09 PM

This is a good example of what predictive analytics are.

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Eins Bernardo
Sent: Wednesday, August 05, 2009 4:33 AM
To: [hidden email]
Subject: gain chart

1. Gain was defined in the book as the proportion of total HITS that occurs in each quantile. What do you mean by "HITS"?

2. What is the interpretation of the point (40,85) on the gain chart? That is, 40% on the horizontal axis and 85% on the vertical axis?

Eins

Adding more friends is quick and easy.
Import them over to Yahoo! Mail today!

Importing contacts has never been easier.
Bring your friends over to Yahoo! Mail today!

Surf faster.
Internet Explorer 8 optmized for Yahoo! auto launches 2 of your favorite pages everytime you open your browser.Get IE8 here! (It's free)

Bring your friends to the fun.
Invite your friends from Hotmail, Gmail to Yahoo! Mail today!

ViAnn Beadle

Re: gain chart

In reply to this post by E. Bernardo

On reflection, please go look at the case study (Help>Case Studies) for Decision Trees. Select the Decision Trees Option>Using Decision Trees to Evaluate Credit Risk.

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Eins Bernardo
Sent: Friday, August 07, 2009 6:39 PM
To: [hidden email]
Subject: Re: gain chart

Sorry for the unclear explanation.

Eins

--- On Thu, 8/6/09, ViAnn Beadle <[hidden email]> wrote:

From: ViAnn Beadle <[hidden email]>
Subject: Re: gain chart
To: [hidden email]
Date: Thursday, 6 August, 2009, 1:16 AM

You’re going to have to give us a reference for your question. What text are you looking at? You are now talking about “VALUES” and points P1, P2, P3—you didn’t mention any of this in your first question.

From: Eins Bernardo [mailto:[hidden email]]
Sent: Wednesday, August 05, 2009 6:42 PM
To: [hidden email]; ViAnn Beadle
Subject: Re: gain chart

ViAnn wrote:

A model was built using some type of statistical method. Usually this is done via CART or CHAID. In SPSS, you would use the TREES command.

I am sorry for the question?

Eins

--- On Wed, 8/5/09, ViAnn Beadle <[hidden email]> wrote:

From: ViAnn Beadle <[hidden email]>
Subject: Re: gain chart
To: [hidden email]
Date: Wednesday, 5 August, 2009, 2:40 PM

A model was built using some type of statistical method. Usually this is done via CART or CHAID. In SPSS, you would use the TREES command.

From: Eins Bernardo [mailto:[hidden email]]
Sent: Wednesday, August 05, 2009 8:30 AM
To: [hidden email]; ViAnn Beadle
Subject: RE: gain chart

Thank you ViAnn for your comment. Follow-up question: How did the customers classified as being in the top 40%?

--- On Wed, 8/5/09, ViAnn Beadle <[hidden email]> wrote:

From: ViAnn Beadle <[hidden email]>
Subject: RE: gain chart
To: "'Eins Bernardo'" <[hidden email]..ph>, [hidden email]
Date: Wednesday, 5 August, 2009, 1:09 PM

This is a good example of what predictive analytics are.

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Eins Bernardo
Sent: Wednesday, August 05, 2009 4:33 AM
To: [hidden email]
Subject: gain chart

1. Gain was defined in the book as the proportion of total HITS that occurs in each quantile. What do you mean by "HITS"?

2. What is the interpretation of the point (40,85) on the gain chart? That is, 40% on the horizontal axis and 85% on the vertical axis?

Eins

Adding more friends is quick and easy.
Import them over to Yahoo! Mail today!

Importing contacts has never been easier.
Bring your friends over to Yahoo! Mail today!

Surf faster.
Internet Explorer 8 optmized for Yahoo! auto launches 2 of your favorite pages everytime you open your browser.Get IE8 here! (It's free)

Bring your friends to the fun.
Invite your friends from Hotmail, Gmail to Yahoo! Mail today!

E. Bernardo

Re: gain chart

In reply to this post by ViAnn Beadle

Thanks ,a lot, ViAnn. I close this thread.

Best,

Best

--- On Sat, 8/8/09, ViAnn Beadle <[hidden email]> wrote:

From: ViAnn Beadle <[hidden email]>
Subject: Re: gain chart
To: [hidden email]
Date: Saturday, 8 August, 2009, 1:09 AM

The variable is the propensity score computed for the target variable in the model. That propensity score identifies the best customers. Amount purchased might be such a target variable but the target variable usually is something like propensity to buy anything at all. This is not the place to discuss all this and I recommend that you read up on the literature.

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Eins Bernardo
Sent: Friday, August 07, 2009 6:39 PM
To: [hidden email]
Subject: Re: gain chart

Sorry for the unclear explanation.

I want to understand the horizontal axis of the gain chart. The horizontal axis is "percentile", right?. How did the percentile computed? To identify the top 40% customers, is it not that the customers are sorted from highest to lowest, then identify the top 40%? What to sort? In this case I am asking what to sort? I think there is one VARIABLE used in order to sort the customers, right?. What is this variable? Sorry, I was using VALUES in my previous email instead of VARIABLE.

To further clarify, suppose the we sort the 100 customers by "amount purchased". The VARIABLE being the basis of sorting the customers is "amount purchased", right?. Once the cusromer is already sorted, we can identify the top 1%, 2%, ,,,40%,,,,99% customers (the percentiles). Clearly, the percentiles were obtained from the "amount purchased" variable. Now the horizontal axis of the gain chart is percentile. I am asking of the variable as basis of getting the percentile.

Eins

--- On Thu, 8/6/09, ViAnn Beadle <[hidden email]> wrote:

From: ViAnn Beadle <[hidden email]>
Subject: Re: gain chart
To: [hidden email]
Date: Thursday, 6 August, 2009, 1:16 AM

You’re going to have to give us a reference for your question. What text are you looking at? You are now talking about “VALUES” and points P1, P2, P3—you didn’t mention any of this in your first question.

From: Eins Bernardo [mailto:[hidden email]]
Sent: Wednesday, August 05, 2009 6:42 PM
To: [hidden email]; ViAnn Beadle
Subject: Re: gain chart

ViAnn wrote:

A model was built using some type of statistical method. Usually this is done via CART or CHAID. In SPSS, you would use the TREES command.

The horizontal line in the chart is quantile. It simply sorts the "VALUES" from highest to lowest, then put some points P1, P2,P3, P4 ....P99 such that the top 40% of the customers fall above P4. Yes I am aware of the classification TREEs. What I didnt understand are the "VALUES"? What are these VALUES? Are these probabilities? Probabilities of what?

I am sorry for the question?

Eins

--- On Wed, 8/5/09, ViAnn Beadle <[hidden email]> wrote:

From: ViAnn Beadle <[hidden email]>
Subject: Re: gain chart
To: [hidden email]
Date: Wednesday, 5 August, 2009, 2:40 PM

A model was built using some type of statistical method. Usually this is done via CART or CHAID. In SPSS, you would use the TREES command.

From: Eins Bernardo [mailto:[hidden email]]
Sent: Wednesday, August 05, 2009 8:30 AM
To: [hidden email]; ViAnn Beadle
Subject: RE: gain chart

Thank you ViAnn for your comment. Follow-up question: How did the customers classified as being in the top 40%?

--- On Wed, 8/5/09, ViAnn Beadle <[hidden email]> wrote:

From: ViAnn Beadle <[hidden email]>
Subject: RE: gain chart
To: "'Eins Bernardo'" <[hidden email]..ph>, [hidden email]
Date: Wednesday, 5 August, 2009, 1:09 PM

Hits is the number or percent of the target action occurring. In this case it was probably is churn. The gains chart plots cumulative gain by decile. So the top 40% of customers accounted for 85% of the customers churning. The more the curve varies from a diagonal line, the better the model is in identifying customers likely to churn.

This is a good example of what predictive analytics are.

From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of Eins Bernardo
Sent: Wednesday, August 05, 2009 4:33 AM
To: [hidden email]
Subject: gain chart

I've read books on predictive models particularly predicting whether a customer churned or not. I saw a gain chart where the horizontal is percentile while the vertical is % gain, but I did not understand. I need your help for the two questions:

1. Gain was defined in the book as the proportion of total HITS that occurs in each quantile. What do you mean by "HITS"?

2. What is the interpretation of the point (40,85) on the gain chart? That is, 40% on the horizontal axis and 85% on the vertical axis?

Eins

Adding more friends is quick and easy.
Import them over to Yahoo! Mail today!

Importing contacts has never been easier.
Bring your friends over to Yahoo! Mail today!

Surf faster.
Internet Explorer 8 optmized for Yahoo! auto launches 2 of your favorite pages everytime you open your browser.Get IE8 here! (It's free)

Bring your friends to the fun.
Invite your friends from Hotmail, Gmail to Yahoo! Mail today!

Cleaner, Better, Faster - Experience the new Faster Yahoo! Mail today!