Subscribe to DSC Newsletter

Hi, I work for a group named Survey Sampling Int.  We build giant online panels and communities of survey respondents - often we give out various types of incentives for survey participation.


I'm looking for assistance and advice on the following 4 categories:


1)      Discovering fraud patterns of people trying to game the system and provide recommendations for how we minimize fraud and analyze behaviors.

2)      Determining the best individualized incentive strategy (much like a casino).  How do we give out rewards in such a way to maximize the yield of people taking surveys?

3)      Optimize allocation of respondents to available surveys in real-time as they land with us to take a survey (router optimization using a predictive model, evaluating all incoming traffic)

4)      Panel behavior and performance overall based on our existing (and future) needs – predictive demand model (insuring that we have enough supply to meet our future demand for survey completes)


Thanks for any info in advance.


Kind regards,


[email protected]

Tags: MR, incentives, marketing, marketingresearch, panels, research, rewards, samples, sampling, surveys

Views: 144

Reply to This

Replies to This Discussion

1) is a very big subject. There are a number of ways to approach it analytically. Here are some of my thoughts

- Start with understanding the underlying process that is resulting in fraud. Are the fraudsters trying to obtaining something illegally or is it users just trying to game the system? An understanding of the process will yield a good understanding of where to get the data the model the fraud behavior
- Probability Scoring. Once you've identified the bad behavior you're going to want to try to predict the probability of that bad behavior. I recommend traditional scoring methods like Logistic Regression and move up to more complicated methods of Neural Nets or other Machine Learning methods. It really depends on what is being modeled to score and how often results are needed.
I may have an answer for you, based on the solution GT data mining that has two relevant qualities: (a) it can find by itself typical patterns of behavior including rare patterns; (b) it defines the relevant combined key factors for each pattern.

It seems that the 4 requirements have in common a similar need to identify effective governing rules as following:

1. FRAUD counter act - focusing on typical fraud patterns of behavior.
2. Effective INCENTIVES - identifying existing patterns that have POSITIVE GROWTH FACTORS.
3. ALLOCATION to surveys - would use patterns' characteristics which include "current traffic" and "load".
4. PREDICTION - with GT rule base, which is more accurate than the commonly used extrapolation of past events.


Where are you standing today with the project?


I think that the 4 categories may interlink. Fraudulent Patterns, Incentives, Resource leveling, and Projections – all benefit from being able to identify the phenomena and key factors of events.


On Data Science Central

© 2019 is a subsidiary and dedicated channel of Data Science Central LLC   Powered by

Badges  |  Report an Issue  |  Privacy Policy  |  Terms of Service