Survey Software, Web Survey, Online Surveys, and Enterprise Feedback Management solutions from Vovici

Your email:
   

Welcome to the Listening Post!

Your single source for everything Voice of the Customer (VoC) and Customer Experience (CxP). And, don’t forget you can follow us on twitter @vovici, or come check us out on Facebook and join the Vovici Network on LinkedIn.

 

Current Articles | RSS Feed RSS Feed

Sample Sourcing Lessons from Panel Professionals

 
river source

Twenty eight of the 30 panel providers we researched in our Lessons from Professional Panel Providers study use non-probability samples, recruiting panelists in an ad-hoc fashion rather than ensuring with a similar probability that any household or individual is invited to join the panel. Such convenience samples are much more affordable than building probability-based panels.

The affordability comes at a cost in terms of what types of research can be done. The AAPOR Online Panel Task Force concludes that non-probability panels should not be used for estimating population values. AAPOR (the American Association for Public Opinion Research) says such panels can be used to contrast the receptivity of different target markets to new product concepts and to determine how attitudes, behavior and intentions interact with the personal attributes of respondents. Such panels are satisfactory for many businesses to use when studying Internet behavior and potential opportunities for Internet services.

Professional panels have either closed or open membership. Many commercial panels are open for anyone to join, typically from the panel provider's web site, while others have closed membership and recruit participants using mailing lists and web partners.

Only closed panels use probability sampling, where the panel provider has randomly targeted households to join the panel using Random Digit Dialing or Address Based Sampling. Because external selection is a key part of probability sampling, such panels typically have extensive recruitment campaigns, starting with direct mail campaigns and culminating in a series of telephone calls. Because of the expense and level of expertise required, few professional panels build panels in this way; Knowledge Networks provides a probability sample of all U.S. households, and Norstat provides a probability sample of all households that have access to the Internet (i.e., "the online population") in the Nordic countries, the Baltic states and Poland.

What Panel Pros Can Teach Us about Sample Sourcing

  1. Align your panel design with the types of decisions you will make and the reliability you need. If you need to estimate the percentage of your customers with certain attitudes or propensities to purchase, then build a probability panel by randomly selecting customers to participate (this is much easier for organizations with a known and documented customer base, such as ecommerce businesses and B2B organizations with large accounts). If you are more interested in qualitative insights (directional but not projectable), then take a more ad-hoc approach to recruiting panelists. 
  2. For non-probability panels, be creative about promoting the panel. Use corporate websites for recruitment, running banner ads across your organization's sites marketing the panel. Include promotional materials for the panel with your online and offline communications to customers.
  3. For B2C firms or other organizations with offline relationships, consider building a panel of loyal customers instead. It is cost prohibitive for traditional retailers, for instance, to build online panels of general customers. Instead of creating a panel to represent the average customer, work with your brand's loyalty program to create a panel of your best customers.

Want to learn more? For the full results of this study, download our white paper, Panel Management Secrets: Lessons from the Professionals, or view our recorded research webinar of the same name.

Comments

Do you think that closed panels results in better sampling because they use probably sampling? Doesn't self-selection/internet availability have a far greater impact on the generalizability of results? I know, i keep harping on this point but I really think that there is no such thing as market research that truly uses and thoroughly follows though with statistical probability sampling.
Posted @ Friday, April 09, 2010 8:22 AM by Annie Pettit
Amend that comment to "almost no panel" and you would be right, Annie. I would encourage you to read the Knowledge Networks' ESOMAR 26 FAQ to see how hard it is to do it right. 
 
E-commerce businesses can do true probability sampling, and a surprising number are. And more businesses are e-commerce than you realize (like low-cost airlines!). Other organizations can do probability sampling of the e-commerce side of their business. The math of probability sampling is solid, the tradition is well established, and modern techniques enable it to be used by firms that could never have afforded to in the past.
Posted @ Friday, April 09, 2010 10:05 AM by Jeffrey Henning
This is a common misperception (shared by the AAPOR report that you mention). You can do "true" probability sampling if you have no nonresponse or attrition. What matters, however, is the sample inclusion probabiity (not the selection probability, unless there is no nonresponse). The fact that an e-commerce site can randomly invite people to a survey does not mean that people randomly respond. The same goes for a panel that uses an initial stage of probabilistic selection. To assume that a sample with a 5% or a 10% response rate is somehow guaranteed to be representative because there is one stage of random selection is silly.
Posted @ Saturday, April 10, 2010 11:35 AM by doug
Thanks, Doug -- please help me provide the right advice to my readers. I agree with your comment about 5-10% response. I advocate reaching at least a 20% response rate if representativity is desired, based on the advice of Holbrook, Krosnick and Pfent, as I write about in Optimize rather than Maximize Response Rates. Second, I would normally say that if you reached that with one invite, great. It sounds like you are saying that one invite does not equal "true" external selection, and that researchers should always use a series of reminders in order to improve representativeness. Let me know the best practices that you recommend. Many thanks!
Posted @ Saturday, April 10, 2010 11:45 AM by Jeffrey Henning
I don't think that is the correct interpretation of the oft-reported "response rates don't matter" results. The two studies by Keeter show little difference between response rates in the 25% range and those in the 50% range. There are, however, substantial differences between either of these and, say, the Current Population Survey (which has a response rate around 90%). The typical media poll (which except for one network that claims a response rate 2-3x higher than the others) now have response rates around 15%. The samples have under-representation of high school dropouts, 18-24 years olds, Hispanics, African-American, and men, and the cross-classification of these cells is truly horrible (with some cells needing weights in excess of 10, i.e. 90% underrepresentation!). The correct interpretation, I think, is that unless nonresponse is very small (less than 10%), it's likely to have some impact regardless of its level. 
 
The problem of selecting site visitors is quite interesting. The IP header provides some information (OS, browser, IP address--which can be geocoded) about all visitors that can be used for stratification. Similarly, page views and other site behavior can be measured. It is also possible to employ balanced sampling methods (see Tille's book on sampling algorithms). 
 
Your blog is always an interesting read, even (especially?) when I disagree.
Posted @ Saturday, April 10, 2010 12:39 PM by doug
Post Comment
Name
 *
Email
 *
Website (optional)
Comment
 *

Allowed tags: <a> link, <b> bold, <i> italics

Latest Posts

Loading
What's New
Don't Be in the 4%
VoC on Twitter
Verint Blog
Verint Blog: Read the Latest from the Verint Systems Blog