Welcome visitor you can log in or create an account

800.275.2827

Consumer Insights. Market Innovation.

blog-page
George Hausser

George Hausser

When not dissecting his opponent’s game on the tennis court, George uses his problem-solving skills to help TRC Research Directors with study design and reporting challenges.

2016 election sample representativenessI always dread the inevitable "What do you do?" question. When you tell someone you are in market research you can typically expect a blank stare or a polite nod; so you must be prepared to offer further explanation. Oh, to be a doctor, lawyer or auto mechanic – no explanation necessary!

Of course, as researchers, we grapple with this issue daily, but it is not often we get to hear it played out on major news networks. After one of the debates, I heard Wolf Blitzer on CNN arguing (yes arguing) with one of the campaign strategists about why the online polls being quoted were not "real" scientific polls. Wolf's point was that because the Internet polls being referenced were from a self-selected sample their results were not representative of the population in question (likely voters). Of course, Wolf was correct, and it made me smile to hear this debated on national TV.

A week or so later I heard an even more, in-depth consideration of the same issue. The story was about how the race was breaking down in key swing states. The poll representative went through the results for key states one-by-one. When she discussed Nevada she raised a red flag as to interpreting the poll (which has one candidate ahead by 2 - % points). She further explained it is difficult to obtain a representative sample in Nevada due to a number of factors (odd work hours, transient population, large Spanish speaking population). Her point was that they try to mitigate these issues, but any results must be viewed with a caveat.

Aside from my personal delight that my day-to-day market research concerns are newsworthy, what is the take-away here? For me, it reinforces how important it is to do everything in our power to ensure that for each study our sample is representative. The advent of online data collection, the proliferation of cell phone use and do-it-yourself survey tools may have made the task more difficult, but no less important. When doing sophisticated conjoint, segmentation or max-diff studies, we need to keep in mind that they are only as good as the sample that feeds them.

Hits: 618 0 Comments

When working with clients on parameters for a conjoint design, there is often an assumption that the design includes a current product configuration, or base case. This base case provides a benchmark against which new configurations can be compared.  
Having a benchmark can be both useful and comforting when analyzing the conjoint results. Replicating a base case allows us to reference important metrics that are known for that product (for example, market share, CPU, revenue, etc.). As we configure new products and compare their appeal to our base case, we can gain insight into how these key metrics might be impacted.
Aside from establishing a benchmark, having a base case is also critical if there is concern about cannibalization.  If the expectation for the new product is that it will compete in the market with a current configuration it is critical to understand what impact the new product will have on the current landscape.
However, allowing for a base case in the conjoint design is not always warranted. As products become more dissimilar from current offerings it can become difficult to include a base case. Trying to integrate the components of a current and new product that don’t share many characteristics can lead to conjoint parameters that are too complex to administer, or create apples to oranges comparisons. It is not wrong to leave out a base case as long as it is understood there will be no benchmark comparison.
One hybrid solution to consider is to allow for a set choice that reads something like “None of these, prefer the PRODUCTS currently available”. This is similar to a typical “none” option in the conjoint but provides a bit more information; specifically, that they would not leave the category but are not interested in the new, very different product configuration. Of course this solution would not be appropriate in all instances but does provide a good compromise.
Ultimately, the extent to which “real products” are modeled with a conjoint study’s parameters is a function of the specific information needs and the complexity of the design. Most of the time we want to include that dose of “reality” in our design but don’t be afraid to leave it behind if warranted.

conjoint analysis design

Hits: 1237 0 Comments

Want to know more?

Give us a few details so we can discuss possible solutions.

Please provide your Name.
Please provide a valid Email.
Please provide your Phone.
Please provide your Comments.
Enter code below : Enter code below :
Please Enter Correct Captcha code
Our Phone Number is 1-800-275-2827
 Find TRC on facebook  Follow us on twitter  Find TRC on LinkedIn

Our Clients