McKernan, Gina
(2018)
CUSTOMER SEGMENTATION APPROACHES: A COMPARISON OF METHODS WITH DATA FROM THE MEDICARE HEALTH OUTCOMES SURVEY.
Doctoral Dissertation, University of Pittsburgh.
(Unpublished)
Abstract
Model-based segmentation approaches are particularly useful in healthcare consumer research, where the primary goal is to identify groups of individuals who share similar attitudinal and behavioral characteristics, in order to develop engagement strategies, create products, and allocates resources tailored to the specific needs of each segment group. Despite the growing research and literature on segmentation models, many healthcare researchers continue to use demographic variables only to classify consumers into groups; while failing to uncover unique patterns, relationships, and latent traits and relationships. The primary aim of this study was to 1) examine the differences in outcomes when classification methods (K-Means and LCA) for segmentation was used in conjunction with continuous and dichotomous scales; and 2) examine the differences in outcomes when prediction methods (CHAID and Neural Networks) for segmentation was used in conjunction with binary and continuous dependent variables and a variation of the classification algorithm. For the purpose of comparison across methods, data from the Medicare Health Outcome Survey was used in all conditions. Results indicated that the best segment class solution was dependent upon both the method and treatment of the inputs and dependent variable for both classification and prediction problems. When the input depression scale was dichotomized, the K-Means model yielded a 6 segment best-class-solution, whereas the LCA model yielded 9 distinct segment classes. On the other hand, LCA models yielded the same segment solution (9 classes), irrespective of the treatment of the depression scale. Similarly, differences in outcomes were identified when the dependent variable was continuous vs. binary when prediction models were used to segment survey respondents. When the outcome was dichotomous, CHAID models resulted in a 5-segment solution, compared to a 6-segment solution for Neural Networks. On the other hand, the binary dependent variable produced a 4-segment solution for both CHAID and Neural Network models. In addition, the interpretation of the segment class profiles is dependent upon both method and condition (input and treatment of dependent variable).
Share
| Citation/Export: |
|
| Social Networking: |
|
Details
| Item Type: |
University of Pittsburgh ETD
|
| Status: |
Unpublished |
| Creators/Authors: |
|
| ETD Committee: |
|
| Date: |
29 January 2018 |
| Date Type: |
Publication |
| Defense Date: |
20 November 2017 |
| Approval Date: |
29 January 2018 |
| Submission Date: |
4 January 2018 |
| Access Restriction: |
No restriction; Release the ETD for access worldwide immediately. |
| Number of Pages: |
231 |
| Institution: |
University of Pittsburgh |
| Schools and Programs: |
School of Education > Psychology in Education |
| Degree: |
PhD - Doctor of Philosophy |
| Thesis Type: |
Doctoral Dissertation |
| Refereed: |
Yes |
| Uncontrolled Keywords: |
segmentation, models, patient, customer, comparison, classification, prediction, latent class, k-means, CHAID, neural networks |
| Date Deposited: |
29 Jan 2018 18:26 |
| Last Modified: |
29 Jan 2018 18:26 |
| URI: |
http://d-scholarship.pitt.edu/id/eprint/33680 |
Metrics
Monthly Views for the past 3 years
Plum Analytics
Actions (login required)
 |
View Item |
![[feed]](/images/feed-icon-32x32.png){kind=icon}