ASA Connect

Hello Everyone,

I'm venturing into new territory and I have a serious bad case of statistical brain block. 

I need to analyze whether or not purchasers of large, expensive commercial equipment switch between lenders to finance each new equipment purchase.  

I have a large dataset comprising multiple unique purchases by a large number of buyers and the cognate lender that financed each purchase.  The data include the type of equipment purchased.  The data range over several years. 

The goal of this study is to help lenders to obtain an understanding of why equipment purchasers may or may not switch between lenders and what influences the borrowers to choose to switch.

I'm having trouble devising a mental picture of how to approach this analysis.  Generally, if I read enough, I can get started, but that approach isn't working on this problem. To complicate matters, I can't find published studies related to this in either the peer-reviewed or industry literature), possibly because I'm not using the correct keywords in my search.

Some of my jumbled thoughts are:

1. Contingency tables to look at associations between borrower and lender, but that doesn't capture borrowing for equipment over time or whether an earlier borrowing decision affects a later borrowing decision.
2. Does a previous borrowing decision affect a later borrowing decision.
3. Does the type of equipment purchased affect the lender decision?
4. Regression analysis to identify important characteristics that affect the lender choice decision.

Would anyone care to offer advice or insight on a problem like this?  I'd appreciate any insight you might be able to offer to break my statistical brain block.

Thanks.

Linda

------------------------------
Linda A. Landon, PhD, ELS, PRC

Research Communiqué
Business, Marketing, & Policy Research

www.researchcommunique.com
LandonPhD@ResearchCommunique.com
573-797-4517

PhD, Molecular Pharmacology
Graduate Certificate, Applied Statistics
Board-Certified Editor in the Life Sciences
------------------------------

Here's one simple idea: since switching lenders is of interest, make a variable that is 1 if the purchaser switched and 0 otherwise. Then, model the probability of switching via logistic regression.

Some pros and cons of this approach:

1. If you have covariates, you can include them to see if they associate with greater or smaller probabilities of switching.
2. If you have just two entries for each purchaser in your dataset, then you could collapse that into one row in your new dataset, with a 1/0 for whether they switched lenders. If there are more than two entries for a purchaser, then you have to decide how to code the "switched" variable (it is 1 if they switched lenders from the first to the second time, but is it 1 or 0 if they switched back to the original lender on their third purchase?)
3. If you have more than two entries for each purchaser, then you might consider including "purchaser" as a random effect. This may not matter very much, but it would capture any correlation among responses within particular purchasers. If taking this route, I'd recommend first working through the simpler model that ignores this possible random effect, but then checking to see if the model improves when accounting for this potential correlation.

Hope this helps.
Byran

------------------------------
Byran Smucker
Associate Professor of Statistics
Miami University
------------------------------

Original Message:
Sent: 02-24-2019 19:15
From: Linda Landon
Subject: Analyzing Switching Business Lenders by Purchasers of Commercial Vehicles

Hello Everyone,

I'm venturing into new territory and I have a serious bad case of statistical brain block.

I need to analyze whether or not purchasers of large, expensive commercial equipment switch between lenders to finance each new equipment purchase.

I have a large dataset comprising multiple unique purchases by a large number of buyers and the cognate lender that financed each purchase.  The data include the type of equipment purchased.  The data range over several years.

The goal of this study is to help lenders to obtain an understanding of why equipment purchasers may or may not switch between lenders and what influences the borrowers to choose to switch.

I'm having trouble devising a mental picture of how to approach this analysis.  Generally, if I read enough, I can get started, but that approach isn't working on this problem. To complicate matters, I can't find published studies related to this in either the peer-reviewed or industry literature), possibly because I'm not using the correct keywords in my search.

Some of my jumbled thoughts are:

1. Contingency tables to look at associations between borrower and lender, but that doesn't capture borrowing for equipment over time or whether an earlier borrowing decision affects a later borrowing decision.
2. Does a previous borrowing decision affect a later borrowing decision.
3. Does the type of equipment purchased affect the lender decision?
4. Regression analysis to identify important characteristics that affect the lender choice decision.

Would anyone care to offer advice or insight on a problem like this?  I'd appreciate any insight you might be able to offer to break my statistical brain block.

Thanks.

Linda

------------------------------
Linda A. Landon, PhD, ELS, PRC

Research Communiqué
Business, Marketing, & Policy Research

www.researchcommunique.com
<maskemail>LandonPhD@...</maskemail>
573-797-4517

PhD, Molecular Pharmacology
Graduate Certificate, Applied Statistics
Board-Certified Editor in the Life Sciences
------------------------------

Bryan,

Thanks for this advice.   I really appreciate that you took the time to share your insight.

The idea of logistic regression had occurred to me but I was unsure whether it would be a valid application exactly because of the points that you raise:  There are multiple entries for each purchaser.  In a few cases, there are hundreds of entries for one purchaser.  

I wonder if it would be possible to view each set of 2 successive transactions as a single observation.   If a purchaser switched lenders between transaction 1 and transaction 2, then that would be coded 1; if not, then it would be coded zero.  If a purchaser switched lenders between transaction  2 and transaction 3, then that would be coded 1; if not, then it would be coded zero.  If a purchaser switched lenders between transaction 3 and transaction 4, then that would be coded 1; if not, then it would be coded zero.  And so on...   A repeated measures analysis, perhaps applying generalized estimating equations, could be applied to these data with the coded response for each new pair of transactions being a repeated measure.  

Thanks again for your help.

Linda


------------------------------
Linda A. Landon, PhD, ELS, PRC

Research Communiqué
Business, Marketing, & Policy Research

www.researchcommunique.com
LandonPhD@ResearchCommunique.com
573-797-4517

PhD, Molecular Pharmacology
Graduate Certificate, Applied Statistics
Board-Certified Editor in the Life Sciences
------------------------------

Original Message:
Sent: 02-25-2019 06:20
From: Byran Smucker
Subject: Analyzing Switching Business Lenders by Purchasers of Commercial Vehicles

Here's one simple idea: since switching lenders is of interest, make a variable that is 1 if the purchaser switched and 0 otherwise. Then, model the probability of switching via logistic regression.

Some pros and cons of this approach:

1. If you have covariates, you can include them to see if they associate with greater or smaller probabilities of switching.
2. If you have just two entries for each purchaser in your dataset, then you could collapse that into one row in your new dataset, with a 1/0 for whether they switched lenders. If there are more than two entries for a purchaser, then you have to decide how to code the "switched" variable (it is 1 if they switched lenders from the first to the second time, but is it 1 or 0 if they switched back to the original lender on their third purchase?)
3. If you have more than two entries for each purchaser, then you might consider including "purchaser" as a random effect. This may not matter very much, but it would capture any correlation among responses within particular purchasers. If taking this route, I'd recommend first working through the simpler model that ignores this possible random effect, but then checking to see if the model improves when accounting for this potential correlation.

Hope this helps.
Byran

------------------------------
Byran Smucker
Associate Professor of Statistics
Miami University

Original Message:
Sent: 02-24-2019 19:15
From: Linda Landon
Subject: Analyzing Switching Business Lenders by Purchasers of Commercial Vehicles

Hello Everyone,

I'm venturing into new territory and I have a serious bad case of statistical brain block.

I need to analyze whether or not purchasers of large, expensive commercial equipment switch between lenders to finance each new equipment purchase.

I have a large dataset comprising multiple unique purchases by a large number of buyers and the cognate lender that financed each purchase.  The data include the type of equipment purchased.  The data range over several years.

The goal of this study is to help lenders to obtain an understanding of why equipment purchasers may or may not switch between lenders and what influences the borrowers to choose to switch.

I'm having trouble devising a mental picture of how to approach this analysis.  Generally, if I read enough, I can get started, but that approach isn't working on this problem. To complicate matters, I can't find published studies related to this in either the peer-reviewed or industry literature), possibly because I'm not using the correct keywords in my search.

Some of my jumbled thoughts are:

1. Contingency tables to look at associations between borrower and lender, but that doesn't capture borrowing for equipment over time or whether an earlier borrowing decision affects a later borrowing decision.
2. Does a previous borrowing decision affect a later borrowing decision.
3. Does the type of equipment purchased affect the lender decision?
4. Regression analysis to identify important characteristics that affect the lender choice decision.

Would anyone care to offer advice or insight on a problem like this?  I'd appreciate any insight you might be able to offer to break my statistical brain block.

Thanks.

Linda

------------------------------
Linda A. Landon, PhD, ELS, PRC

Research Communiqué
Business, Marketing, & Policy Research

www.researchcommunique.com
<maskemail>LandonPhD@...</maskemail>
573-797-4517

PhD, Molecular Pharmacology
Graduate Certificate, Applied Statistics
Board-Certified Editor in the Life Sciences
------------------------------

Yes, Linda, I think using repeated measures in the way you've suggested is a reasonable approach.

Good luck!

Byran

------------------------------
Byran Smucker
Associate Professor of Statistics
Miami University
------------------------------

Original Message:
Sent: 02-25-2019 20:47
From: Linda Landon
Subject: Analyzing Switching Business Lenders by Purchasers of Commercial Vehicles

Bryan,

Thanks for this advice.   I really appreciate that you took the time to share your insight.

The idea of logistic regression had occurred to me but I was unsure whether it would be a valid application exactly because of the points that you raise:  There are multiple entries for each purchaser.  In a few cases, there are hundreds of entries for one purchaser.

I wonder if it would be possible to view each set of 2 successive transactions as a single observation.   If a purchaser switched lenders between transaction 1 and transaction 2, then that would be coded 1; if not, then it would be coded zero.  If a purchaser switched lenders between transaction  2 and transaction 3, then that would be coded 1; if not, then it would be coded zero.  If a purchaser switched lenders between transaction 3 and transaction 4, then that would be coded 1; if not, then it would be coded zero.  And so on...   A repeated measures analysis, perhaps applying generalized estimating equations, could be applied to these data with the coded response for each new pair of transactions being a repeated measure.

Thanks again for your help.

Linda


------------------------------
Linda A. Landon, PhD, ELS, PRC

Research Communiqué
Business, Marketing, & Policy Research

www.researchcommunique.com
<maskemail>LandonPhD@...</maskemail>
573-797-4517

PhD, Molecular Pharmacology
Graduate Certificate, Applied Statistics
Board-Certified Editor in the Life Sciences

Original Message:
Sent: 02-25-2019 06:20
From: Byran Smucker
Subject: Analyzing Switching Business Lenders by Purchasers of Commercial Vehicles

Here's one simple idea: since switching lenders is of interest, make a variable that is 1 if the purchaser switched and 0 otherwise. Then, model the probability of switching via logistic regression.

Some pros and cons of this approach:

1. If you have covariates, you can include them to see if they associate with greater or smaller probabilities of switching.
2. If you have just two entries for each purchaser in your dataset, then you could collapse that into one row in your new dataset, with a 1/0 for whether they switched lenders. If there are more than two entries for a purchaser, then you have to decide how to code the "switched" variable (it is 1 if they switched lenders from the first to the second time, but is it 1 or 0 if they switched back to the original lender on their third purchase?)
3. If you have more than two entries for each purchaser, then you might consider including "purchaser" as a random effect. This may not matter very much, but it would capture any correlation among responses within particular purchasers. If taking this route, I'd recommend first working through the simpler model that ignores this possible random effect, but then checking to see if the model improves when accounting for this potential correlation.

Hope this helps.
Byran

------------------------------
Byran Smucker
Associate Professor of Statistics
Miami University

Original Message:
Sent: 02-24-2019 19:15
From: Linda Landon
Subject: Analyzing Switching Business Lenders by Purchasers of Commercial Vehicles

Hello Everyone,

I'm venturing into new territory and I have a serious bad case of statistical brain block.

I need to analyze whether or not purchasers of large, expensive commercial equipment switch between lenders to finance each new equipment purchase.

I have a large dataset comprising multiple unique purchases by a large number of buyers and the cognate lender that financed each purchase.  The data include the type of equipment purchased.  The data range over several years.

The goal of this study is to help lenders to obtain an understanding of why equipment purchasers may or may not switch between lenders and what influences the borrowers to choose to switch.

I'm having trouble devising a mental picture of how to approach this analysis.  Generally, if I read enough, I can get started, but that approach isn't working on this problem. To complicate matters, I can't find published studies related to this in either the peer-reviewed or industry literature), possibly because I'm not using the correct keywords in my search.

Some of my jumbled thoughts are:

1. Contingency tables to look at associations between borrower and lender, but that doesn't capture borrowing for equipment over time or whether an earlier borrowing decision affects a later borrowing decision.
2. Does a previous borrowing decision affect a later borrowing decision.
3. Does the type of equipment purchased affect the lender decision?
4. Regression analysis to identify important characteristics that affect the lender choice decision.

Would anyone care to offer advice or insight on a problem like this?  I'd appreciate any insight you might be able to offer to break my statistical brain block.

Thanks.

Linda

------------------------------
Linda A. Landon, PhD, ELS, PRC

Research Communiqué
Business, Marketing, & Policy Research

www.researchcommunique.com
<maskemail>LandonPhD@...</maskemail>
573-797-4517

PhD, Molecular Pharmacology
Graduate Certificate, Applied Statistics
Board-Certified Editor in the Life Sciences
------------------------------

Thanks, Bryan!

------------------------------
Linda A. Landon, PhD, ELS, PRC

Research Communiqué
Business, Marketing, & Policy Research

www.researchcommunique.com
LandonPhD@ResearchCommunique.com
573-797-4517

PhD, Molecular Pharmacology
Graduate Certificate, Applied Statistics
Board-Certified Editor in the Life Sciences
------------------------------

Original Message:
Sent: 02-26-2019 14:02
From: Byran Smucker
Subject: Analyzing Switching Business Lenders by Purchasers of Commercial Vehicles

Yes, Linda, I think using repeated measures in the way you've suggested is a reasonable approach.

Good luck!

Byran

------------------------------
Byran Smucker
Associate Professor of Statistics
Miami University

Original Message:
Sent: 02-25-2019 20:47
From: Linda Landon
Subject: Analyzing Switching Business Lenders by Purchasers of Commercial Vehicles

Bryan,

Thanks for this advice.   I really appreciate that you took the time to share your insight.

The idea of logistic regression had occurred to me but I was unsure whether it would be a valid application exactly because of the points that you raise:  There are multiple entries for each purchaser.  In a few cases, there are hundreds of entries for one purchaser.

I wonder if it would be possible to view each set of 2 successive transactions as a single observation.   If a purchaser switched lenders between transaction 1 and transaction 2, then that would be coded 1; if not, then it would be coded zero.  If a purchaser switched lenders between transaction  2 and transaction 3, then that would be coded 1; if not, then it would be coded zero.  If a purchaser switched lenders between transaction 3 and transaction 4, then that would be coded 1; if not, then it would be coded zero.  And so on...   A repeated measures analysis, perhaps applying generalized estimating equations, could be applied to these data with the coded response for each new pair of transactions being a repeated measure.

Thanks again for your help.

Linda


------------------------------
Linda A. Landon, PhD, ELS, PRC

Research Communiqué
Business, Marketing, & Policy Research

www.researchcommunique.com
<maskemail>LandonPhD@...</maskemail>
573-797-4517

PhD, Molecular Pharmacology
Graduate Certificate, Applied Statistics
Board-Certified Editor in the Life Sciences

Original Message:
Sent: 02-25-2019 06:20
From: Byran Smucker
Subject: Analyzing Switching Business Lenders by Purchasers of Commercial Vehicles

Here's one simple idea: since switching lenders is of interest, make a variable that is 1 if the purchaser switched and 0 otherwise. Then, model the probability of switching via logistic regression.

Some pros and cons of this approach:

1. If you have covariates, you can include them to see if they associate with greater or smaller probabilities of switching.
2. If you have just two entries for each purchaser in your dataset, then you could collapse that into one row in your new dataset, with a 1/0 for whether they switched lenders. If there are more than two entries for a purchaser, then you have to decide how to code the "switched" variable (it is 1 if they switched lenders from the first to the second time, but is it 1 or 0 if they switched back to the original lender on their third purchase?)
3. If you have more than two entries for each purchaser, then you might consider including "purchaser" as a random effect. This may not matter very much, but it would capture any correlation among responses within particular purchasers. If taking this route, I'd recommend first working through the simpler model that ignores this possible random effect, but then checking to see if the model improves when accounting for this potential correlation.

Hope this helps.
Byran

------------------------------
Byran Smucker
Associate Professor of Statistics
Miami University

Original Message:
Sent: 02-24-2019 19:15
From: Linda Landon
Subject: Analyzing Switching Business Lenders by Purchasers of Commercial Vehicles

Hello Everyone,

I'm venturing into new territory and I have a serious bad case of statistical brain block.

I need to analyze whether or not purchasers of large, expensive commercial equipment switch between lenders to finance each new equipment purchase.

I have a large dataset comprising multiple unique purchases by a large number of buyers and the cognate lender that financed each purchase.  The data include the type of equipment purchased.  The data range over several years.

The goal of this study is to help lenders to obtain an understanding of why equipment purchasers may or may not switch between lenders and what influences the borrowers to choose to switch.

I'm having trouble devising a mental picture of how to approach this analysis.  Generally, if I read enough, I can get started, but that approach isn't working on this problem. To complicate matters, I can't find published studies related to this in either the peer-reviewed or industry literature), possibly because I'm not using the correct keywords in my search.

Some of my jumbled thoughts are:

1. Contingency tables to look at associations between borrower and lender, but that doesn't capture borrowing for equipment over time or whether an earlier borrowing decision affects a later borrowing decision.
2. Does a previous borrowing decision affect a later borrowing decision.
3. Does the type of equipment purchased affect the lender decision?
4. Regression analysis to identify important characteristics that affect the lender choice decision.

Would anyone care to offer advice or insight on a problem like this?  I'd appreciate any insight you might be able to offer to break my statistical brain block.

Thanks.

Linda

------------------------------
Linda A. Landon, PhD, ELS, PRC

Research Communiqué
Business, Marketing, & Policy Research

www.researchcommunique.com
<maskemail>LandonPhD@...</maskemail>
573-797-4517

PhD, Molecular Pharmacology
Graduate Certificate, Applied Statistics
Board-Certified Editor in the Life Sciences
------------------------------

Linda,

First let me state a caveat: I am not intimately connected to this data, nor the question in depth. I have not analyzes the approach nor the data. My response is from a purely theoretical direction.

W. Edwards Deming published an article in 1975 concerning the difference between enumerative study, and analytic study. Through this article he shows with some mathematical clarity that no matter how accurately we count something, we cannot say anything at all about what 'caused' the count.

Article:
https://pdfs.semanticscholar.org/adec/f8c3cc38faec3e11370561e13d89e7499452.pdf

Semanticscholar		remove preview
View this on Semanticscholar >

Enumerative vs. Analytic wiki
https://en.wikipedia.org/wiki/Analytic_and_enumerative_statistical_studies

Your phrasing of the question, if I read it correctly, shows why you are having difficulty approaching this question. There is no evidence here of a causal nature. No one has been asked 'why' they made a certain choice.

So, it may be that the best response will be to say that the data does not include any sort of analysis, (of a non-enumerative nature), and thus cannot provide any analysis of 'reasons' for making of the decisions in question.

-Terry Rosen

------------------------------
Terry Rosen
BoulderTEC High School
------------------------------

Original Message:
Sent: 02-24-2019 19:15
From: Linda Landon
Subject: Analyzing Switching Business Lenders by Purchasers of Commercial Vehicles

Hello Everyone,

I'm venturing into new territory and I have a serious bad case of statistical brain block.

I need to analyze whether or not purchasers of large, expensive commercial equipment switch between lenders to finance each new equipment purchase.

I have a large dataset comprising multiple unique purchases by a large number of buyers and the cognate lender that financed each purchase.  The data include the type of equipment purchased.  The data range over several years.

The goal of this study is to help lenders to obtain an understanding of why equipment purchasers may or may not switch between lenders and what influences the borrowers to choose to switch.

I'm having trouble devising a mental picture of how to approach this analysis.  Generally, if I read enough, I can get started, but that approach isn't working on this problem. To complicate matters, I can't find published studies related to this in either the peer-reviewed or industry literature), possibly because I'm not using the correct keywords in my search.

Some of my jumbled thoughts are:

1. Contingency tables to look at associations between borrower and lender, but that doesn't capture borrowing for equipment over time or whether an earlier borrowing decision affects a later borrowing decision.
2. Does a previous borrowing decision affect a later borrowing decision.
3. Does the type of equipment purchased affect the lender decision?
4. Regression analysis to identify important characteristics that affect the lender choice decision.

Would anyone care to offer advice or insight on a problem like this?  I'd appreciate any insight you might be able to offer to break my statistical brain block.

Thanks.

Linda

------------------------------
Linda A. Landon, PhD, ELS, PRC

Research Communiqué
Business, Marketing, & Policy Research

www.researchcommunique.com
<maskemail>LandonPhD@...</maskemail>
573-797-4517

PhD, Molecular Pharmacology
Graduate Certificate, Applied Statistics
Board-Certified Editor in the Life Sciences
------------------------------

Hello Terry,

I absolutely agree that this analysis can't be causal; at the very "best", it will be associational.   We are still very early in planning how to best use (or not to use) the available data and whether we need to supplement the variables present in our dataset with additional exogenous variables.  

Slowly, through your advice and the advice of others, I am obtaining a handle on this analysis and breaking through my statistical brain block.  Part of the problem is the inability of our client to state what they want; part of the problem was my statistical brain block.  Right now, I'm somewhat chagrined that the possible paths forward were so obvious but I couldn't seem to see the forest for the trees.

I appreciate having the Demming article.  I had heard of this article but never had a copy of it previously.  Thanks for sharing it with me.  

Thanks so much for taking the time to respond.  I deeply appreciate your help.

Linda

------------------------------
Linda A. Landon, PhD, ELS, PRC

Research Communiqué
Business, Marketing, & Policy Research

www.researchcommunique.com
LandonPhD@ResearchCommunique.com
573-797-4517

PhD, Molecular Pharmacology
Graduate Certificate, Applied Statistics
Board-Certified Editor in the Life Sciences
------------------------------

Original Message:
Sent: 02-26-2019 10:59
From: Terry Rosen
Subject: Analyzing Switching Business Lenders by Purchasers of Commercial Vehicles

Linda,

First let me state a caveat: I am not intimately connected to this data, nor the question in depth. I have not analyzes the approach nor the data. My response is from a purely theoretical direction.

W. Edwards Deming published an article in 1975 concerning the difference between enumerative study, and analytic study. Through this article he shows with some mathematical clarity that no matter how accurately we count something, we cannot say anything at all about what 'caused' the count.

Article:
https://pdfs.semanticscholar.org/adec/f8c3cc38faec3e11370561e13d89e7499452.pdf

Semanticscholar		remove preview
View this on Semanticscholar >

Enumerative vs. Analytic wiki
https://en.wikipedia.org/wiki/Analytic_and_enumerative_statistical_studies

Your phrasing of the question, if I read it correctly, shows why you are having difficulty approaching this question. There is no evidence here of a causal nature. No one has been asked 'why' they made a certain choice.

So, it may be that the best response will be to say that the data does not include any sort of analysis, (of a non-enumerative nature), and thus cannot provide any analysis of 'reasons' for making of the decisions in question.

-Terry Rosen

------------------------------
Terry Rosen
BoulderTEC High School

Original Message:
Sent: 02-24-2019 19:15
From: Linda Landon
Subject: Analyzing Switching Business Lenders by Purchasers of Commercial Vehicles

Hello Everyone,

I'm venturing into new territory and I have a serious bad case of statistical brain block.

I need to analyze whether or not purchasers of large, expensive commercial equipment switch between lenders to finance each new equipment purchase.

I have a large dataset comprising multiple unique purchases by a large number of buyers and the cognate lender that financed each purchase.  The data include the type of equipment purchased.  The data range over several years.

The goal of this study is to help lenders to obtain an understanding of why equipment purchasers may or may not switch between lenders and what influences the borrowers to choose to switch.

I'm having trouble devising a mental picture of how to approach this analysis.  Generally, if I read enough, I can get started, but that approach isn't working on this problem. To complicate matters, I can't find published studies related to this in either the peer-reviewed or industry literature), possibly because I'm not using the correct keywords in my search.

Some of my jumbled thoughts are:

1. Contingency tables to look at associations between borrower and lender, but that doesn't capture borrowing for equipment over time or whether an earlier borrowing decision affects a later borrowing decision.
2. Does a previous borrowing decision affect a later borrowing decision.
3. Does the type of equipment purchased affect the lender decision?
4. Regression analysis to identify important characteristics that affect the lender choice decision.

Would anyone care to offer advice or insight on a problem like this?  I'd appreciate any insight you might be able to offer to break my statistical brain block.

Thanks.

Linda

------------------------------
Linda A. Landon, PhD, ELS, PRC

Research Communiqué
Business, Marketing, & Policy Research

www.researchcommunique.com
<maskemail>LandonPhD@...</maskemail>
573-797-4517

PhD, Molecular Pharmacology
Graduate Certificate, Applied Statistics
Board-Certified Editor in the Life Sciences
------------------------------

Original Message------

Hello Terry,

I absolutely agree that this analysis can't be causal; at the very "best", it will be associational.   We are still very early in planning how to best use (or not to use) the available data and whether we need to supplement the variables present in our dataset with additional exogenous variables.  

Slowly, through your advice and the advice of others, I am obtaining a handle on this analysis and breaking through my statistical brain block.  Part of the problem is the inability of our client to state what they want; part of the problem was my statistical brain block.  Right now, I'm somewhat chagrined that the possible paths forward were so obvious but I couldn't seem to see the forest for the trees.

I appreciate having the Demming article.  I had heard of this article but never had a copy of it previously.  Thanks for sharing it with me.  

Thanks so much for taking the time to respond.  I deeply appreciate your help.

Linda

------------------------------
Linda A. Landon, PhD, ELS, PRC

Research Communiqué
Business, Marketing, & Policy Research

www.researchcommunique.com
LandonPhD@ResearchCommunique.com
573-797-4517

PhD, Molecular Pharmacology
Graduate Certificate, Applied Statistics
Board-Certified Editor in the Life Sciences
------------------------------

Hi Linda

As a CPA looking at the issue it seems to me there are only  a small number of potential predictor variables that could possibly
enter into a borrowers decision to switch lenders:

1. Interest rates. Lower is better for the borrower.
2. Allowable amortization periods. Longer is better for the borrower.
3. Loan to value ratios. The higher the allowed ratio is better for the borrower.

Without even running a regression analysis you can rely on the above variables as the major drivers of borrower decisions.

Given that the dependent variable is discrete "change" "don't change" it looks like you would have to use some form of logistic regression.

These three predictor variables will not usually be independent. Most lenders will have a means of classifying borrowers based on overall credit worthiness. Those with higher overall credit scores will have access to lower borrowing rates, longer amortization periods and higher loan to value ratios. It is highly unlikely that each lender will have a one sized set of lending criteria for all borrowers. This being the case data should be stratified based on the lending standards for those borrowers with high credit scores vs. the standards applied to borrowers with low credit scores.

I hope this helps thinking about the problem.


Mike Elmaleh

------------------------------
Michael Sack Elmaleh
Principal
Michael Sack Elmaleh CPA, CVA
------------------------------

Original Message:
Sent: 02-24-2019 19:15
From: Linda Landon
Subject: Analyzing Switching Business Lenders by Purchasers of Commercial Vehicles

Hello Everyone,

I'm venturing into new territory and I have a serious bad case of statistical brain block.

I need to analyze whether or not purchasers of large, expensive commercial equipment switch between lenders to finance each new equipment purchase.

I have a large dataset comprising multiple unique purchases by a large number of buyers and the cognate lender that financed each purchase.  The data include the type of equipment purchased.  The data range over several years.

The goal of this study is to help lenders to obtain an understanding of why equipment purchasers may or may not switch between lenders and what influences the borrowers to choose to switch.

I'm having trouble devising a mental picture of how to approach this analysis.  Generally, if I read enough, I can get started, but that approach isn't working on this problem. To complicate matters, I can't find published studies related to this in either the peer-reviewed or industry literature), possibly because I'm not using the correct keywords in my search.

Some of my jumbled thoughts are:

1. Contingency tables to look at associations between borrower and lender, but that doesn't capture borrowing for equipment over time or whether an earlier borrowing decision affects a later borrowing decision.
2. Does a previous borrowing decision affect a later borrowing decision.
3. Does the type of equipment purchased affect the lender decision?
4. Regression analysis to identify important characteristics that affect the lender choice decision.

Would anyone care to offer advice or insight on a problem like this?  I'd appreciate any insight you might be able to offer to break my statistical brain block.

Thanks.

Linda

------------------------------
Linda A. Landon, PhD, ELS, PRC

Research Communiqué
Business, Marketing, & Policy Research

www.researchcommunique.com
<maskemail>LandonPhD@...</maskemail>
573-797-4517

PhD, Molecular Pharmacology
Graduate Certificate, Applied Statistics
Board-Certified Editor in the Life Sciences
------------------------------

Hello Mike,

Thanks for your feedback.   Identifying these prospective predictor variables and your discusion of their importance is invaluable and I deeply appreciate that you shared it with me.

I agree that logistic regression is a good candidate analysis.  In particular, using generalized estimating equations to conduct logistic regression has attractive possibilities. 

There was another interesting suggestion that I'm not sure appeared in the online answers:   Average run length to gauge borrower loyalty.  

Thanks for taking the time to answer my question.  I appreciate that you did so.

Linda

------------------------------
Linda A. Landon, PhD, ELS, PRC

Research Communiqué
Business, Marketing, & Policy Research

www.researchcommunique.com
LandonPhD@ResearchCommunique.com
573-797-4517

PhD, Molecular Pharmacology
Graduate Certificate, Applied Statistics
Board-Certified Editor in the Life Sciences
------------------------------

Original Message:
Sent: 02-26-2019 11:05
From: Michael Sack Elmaleh
Subject: Analyzing Switching Business Lenders by Purchasers of Commercial Vehicles

Hi Linda

As a CPA looking at the issue it seems to me there are only  a small number of potential predictor variables that could possibly
enter into a borrowers decision to switch lenders:

1. Interest rates. Lower is better for the borrower.
2. Allowable amortization periods. Longer is better for the borrower.
3. Loan to value ratios. The higher the allowed ratio is better for the borrower.

Without even running a regression analysis you can rely on the above variables as the major drivers of borrower decisions.

Given that the dependent variable is discrete "change" "don't change" it looks like you would have to use some form of logistic regression.

These three predictor variables will not usually be independent. Most lenders will have a means of classifying borrowers based on overall credit worthiness. Those with higher overall credit scores will have access to lower borrowing rates, longer amortization periods and higher loan to value ratios. It is highly unlikely that each lender will have a one sized set of lending criteria for all borrowers. This being the case data should be stratified based on the lending standards for those borrowers with high credit scores vs. the standards applied to borrowers with low credit scores.

I hope this helps thinking about the problem.


Mike Elmaleh

------------------------------
Michael Sack Elmaleh
Principal
Michael Sack Elmaleh CPA, CVA

Original Message:
Sent: 02-24-2019 19:15
From: Linda Landon
Subject: Analyzing Switching Business Lenders by Purchasers of Commercial Vehicles

Hello Everyone,

I'm venturing into new territory and I have a serious bad case of statistical brain block.

I need to analyze whether or not purchasers of large, expensive commercial equipment switch between lenders to finance each new equipment purchase.

I have a large dataset comprising multiple unique purchases by a large number of buyers and the cognate lender that financed each purchase.  The data include the type of equipment purchased.  The data range over several years.

The goal of this study is to help lenders to obtain an understanding of why equipment purchasers may or may not switch between lenders and what influences the borrowers to choose to switch.

I'm having trouble devising a mental picture of how to approach this analysis.  Generally, if I read enough, I can get started, but that approach isn't working on this problem. To complicate matters, I can't find published studies related to this in either the peer-reviewed or industry literature), possibly because I'm not using the correct keywords in my search.

Some of my jumbled thoughts are:

1. Contingency tables to look at associations between borrower and lender, but that doesn't capture borrowing for equipment over time or whether an earlier borrowing decision affects a later borrowing decision.
2. Does a previous borrowing decision affect a later borrowing decision.
3. Does the type of equipment purchased affect the lender decision?
4. Regression analysis to identify important characteristics that affect the lender choice decision.

Would anyone care to offer advice or insight on a problem like this?  I'd appreciate any insight you might be able to offer to break my statistical brain block.

Thanks.

Linda

------------------------------
Linda A. Landon, PhD, ELS, PRC

Research Communiqué
Business, Marketing, & Policy Research

www.researchcommunique.com
<maskemail>LandonPhD@...</maskemail>
573-797-4517

PhD, Molecular Pharmacology
Graduate Certificate, Applied Statistics
Board-Certified Editor in the Life Sciences
------------------------------