One billion and growing

Editor’s note: Kira Signer is a former senior project director, and Andy Korman is former CEO, of Opinionsite Research, New York. 

The world Internet population is currently estimated at .97 to 1.1 billion users1,2,3 – almost one-fifth of the world population overall. In the United States  alone, with Internet penetration at 68 percent (spanning every socioeconomic, educational and ethnic divide)3, it seems that the Web has arrived – specifically as a research tool as well as a mass-market technology woven into every element of our society.

Sampling for online research in the U.S.

After years of discussion, validation research and proving-out of online research conclusions in the business world, most researchers are willing to accept that online research can, under appropriate conditions and restrictions, accurately represent U.S. consumers as a whole 4,5,6,7. Several key factors usually cooperate to argue that online panels (lists of Internet users who have agreed to participate in surveys) are sufficiently representative for good research.

Online consumers can now represent almost the entirety of the U.S. market. With the Internet following an accelerated but otherwise typical innovation-diffusion product cycle8, the classic Rogers curve (accepted by sociological and business researchers)9 is the model most widely applied to illustrate the diffusion of Internet penetration. This model proposes that the bulk of the market is composed of innovators (2-3 percent), early adopters (12-13 percent), early market (34 percent) and late market (34 percent) mainstream consumers, so according to the Rogers curve, at 68 percent U.S. Internet penetration has reached well into the late-market mainstream. The only consumers disproportionately excluded from online representation are thus the late-market “laggards” or “dinosaurs” – consumers who tend to resist purchase until they find it absolutely necessary, or who “are, well, never going to adopt.”10 This consumer rarely represents an enthusiastic consumer or a key target for most companies. For new products, especially novelty products or those that represent an innovation, the opinions of these late adopters may have even less direct impact on product success.

These laggards are disproportionately likely to be older, low-income consumers (aged 65+ with household income less than $30,000) who are unlikely to represent most companies’ target consumers.11 However, penetration of the Internet among users 65 and older increased to 26 percent in 2005, suggesting that penetration will continue to increase.12

Internet panels are now sufficiently large to represent a notable proportion of the underlying population. A large pool of respondents reduces the threat of bias from systematic sampling13. If samples are selected carefully from a large base panel, a wide range of attitudes, opinions and experiences are equally likely to be included. Panels need only to be large enough to counteract low representation of any specific demographic subgroups. Even older, low-income consumers can be well-represented within online panels; though only 2-5 percent of a typical online panel will fit the above definition of a laggard, with member lists varying from 150,000 to 1 million, panel companies can generally access 3,000 – 50,000 individuals meeting this definition.

Even as Internet panels become more representative, telephone surveys and other trusted methodologies become less so. More consumers are abandoning landlines for cell phones (landline penetration has dropped 5 percent in the past 8 years, at an accelerating rate)14, rendering them unavailable for random telephone surveying. As telemarketing, over-surveying and push polls increase respondent frustration, and answering machines complicate respondent contact, industry participation rates are declining15,16. One recent study shows telephone refusal rates up to 75 percent17 among the sub-group of potential respondents who actually answered the phone. Further, while recent “do not call” legislation in the U.S. does not apply to surveying, respondents may not know of the exception, and can assume they should not be contacted by any firm via telephone – with such resistance pushing refusal rates even higher.

The online population in the U.S. skews wealthier, more educated, and more Caucasian than the U.S. population as a whole. In some cases, these targets are more preferred for market research sampling to begin with. For more mainstream surveys, appropriate sampling techniques will correct for this. Careful and proportionate stratified sampling will result in a survey sample that reflects the U.S. as a whole instead of the Internet or panel demographic. Demographic representation in a survey sample depends more on the panelists who were invited and respond, not the overall panel composition. In many cases, research companies can ensure appropriate representation by inviting low-incidence populations disproportionately to the survey. Instead of making the target’s share of invitations proportionate to the panel or the Internet user, specific low-response strata are over-represented to ensure that they will respond at a rate that will reflect their proportion in the underlying population as a whole.

In the U.S., Internet users are “more socially connected than non-users, have a stronger sense of efficacy (perceived control over one’s life), and consume more media (including newspapers, TV, mobile phone usage, etc.) than non-users.”8 This suggests that Internet targets might have more developed and articulate opinions than non-Internet targets. Anecdotal experience with online surveys reveal a wide quality range in open-ended responses – some questions can be extensive and very rich with thoughtful information. Further, since they are heavier consumers of all forms of media8 – including print and television – than non-users, Internet users would be more likely to be consumers of advertising.

Sampling for online research internationally

Clearly, there are many factors influencing whether valid online research can be reliably collected. The accompanying tables synthesize and summarize relevant information on selected countries, in order to provide a recommendation as to the viability of online research in each country.

Internet penetration can vary widely by country – from 2 percent to 78 percent1.  Some may argue that the efficacy of online research is limited to countries with mature markets for Internet access. However, in many countries with moderate to minimal penetration, there is a sound rationale for believing Internet research to be representative.

  • Demographically representative.

Globally, Internet users are typically more likely to be younger and higher-income18,19. For example, Internet penetration in Italy skews low on older, lower-income and Northern respondents (see tables). In many cases, research companies can ensure appropriate representation by careful and proportionate stratified sampling just as easily abroad as they can here in the U.S.

  • Representing the mass market

Internet penetration in some countries is recent, limited, and still accelerating. Following the typical pattern of the technology adoption curve, online panels in some of these countries will represent early adopters predominately or disproportionately. It may be less of a concern in some countries with low and/or recent Internet penetration, if the Internet has been made broadly available across socioeconomic strata. For example, Brazil has followed a program to distribute free Internet access and training to people of all incomes (see tables). In other countries, such as India, recent penetration and a preponderance of early adopters can lead to exaggeratedly positive findings through online research, if early adopters are more positive about the survey topic than the overall population.

Some countries with incomplete Internet penetration have been pursuing connectivity for decades; in countries like Spain or Thailand, barriers to expanding infrastructure limit the market more than adoption diffusion (see tables). Since the Internet in these countries has been available for a relatively long time in well-connected urban areas, late-market consumers can be well-represented in surveys by ensuring high participation of urban respondents – who are in any case more likely to be close to available channels for companies’ products.

For surveys that target the opinions of early adopters, a skew to this target is ideal. In countries where Internet penetration is still a new and evolving phenomenon, this natural skew can be an advantage for some survey topics. For more mainstream topics, several steps can and should be taken to minimize the impact of early adopters on study results for more mass-market surveys. First, surveys should include both recent and long-term users. Second, results from studies in early-adopter countries suggest that new Internet users can report disproportionately positive answers to surveys. Intelligent questionnaire design can help minimize the impact of this effect.

  • Representative of market targets

In many countries, even though few may be online, those who are online represent a much larger proportion of consumer targets than those who are offline. For example, Internet penetration in China is extremely urbanized, and represents fewer than 10 percent of the population overall (see tables) – but since only one-third of the country lives in areas with populations of at least 2,50020, online penetration represents roughly one-quarter of those who shop in stores.  Any company targeting potential retail customers can thus effectively consider China to be a country with moderate Internet penetration of their target customer.

Notes

1 Miniwatts International (2005-2006), Internet World Statistics, published online.
http://www.internetworldstats.com/top25.htm

2 Dagan, R. (2006) Internet Press Releases compiled from ITU, Nielsen, CyberAtlas and other sources, published
online. http://www.dagandesigns.com/perspectives.php

3 Nielsen NetRatings (2006). http://www.nielsen-netratings.com/

4 WebSurveyor (2003), internal data and interview with Canadian researchers, published online. http://www.websurveyor.com/pdf/WebvsMail.pdf 

5 Azar, B. (2000), “A Web of Research,” Monitor on Psychology, v.31:4, American Psychology Association.
http://www.apa.org/monitor/apr00/research.html

6 Perseus (2004), White Papers, published online. http://www.perseus.com/surveytips/thw_websurveys.html

7 Crandall, B. (2000) Decision Analyst white papers. http://www.decisionanalyst.com/publ_art/res.asp

8 Chen, W. and Wellman, B. (2003), Charting and Bridging Digital Divides, special report for the AMD Global
Consumer Advisory Board.  http://www.amd.com/usen/assets/content_type/DownloadableAssets/FINAL_REPORT_CHARTING_DIGI_DIVIDES.pdf

9 Rogers, Everett (1962), Diffusion of Innovations. Free Press. http://www.valuebasedmanagement.net/methods_rogers_innovation_adoption_curve.html

10 Rifkin (2003), “Why New Processes Are Not Adopted,” Advances in Computers, v. 59. http://www.master-systems.com/filecabinet/WhyNewProcessesAreNotAdopted12.pdf

11 Cooper, M. (2000), Disconnected, Disadvantaged and Disenfanchised, Consumers Union/Consumer Reports.
http://www.consumersunion.org/pdf/disconnect.pdf

12 WebMetro (2005b), press release quoting the Pew Internet and American Life Project, published online.
http://www.webmetro.com/news1detail1.asp?id=1208

13 Lane, D.M. (1993), Hyperstat Online Statistics Textbook. http://www.davidmlane.com/hyperstat/

14 United States Telecom Association (undated) http://www.usta.org/

15 O’Rourke, D. et al. (1998), “An Inquiry into Declining RDD Response Rates”. Survey Research, v.29:2.
http://www.srl.uic.edu/publist/Srvrsch/98v29n2.pdf

16 GSMFC Newsletter, 2005. http://www.gsmfc.org/pubs/FIN/newsletter2005.pdf

17 MRA (2004), Alert! Trade Magazine, October 2004 issue. http://www.mra-net.org/publications/alert_pdf/0410.pdf

18 ITU World Communication Development Report, 2003. http://www.itu.int/ITUD/ict/publications/wtdr_03/material/WTDR2003Sum_e.pdf

19 Clickz Press Release (2002), quoting Nielsen/NetRatings, published online.
http://www.clickz.com/stats/sectors/demographics/article.php/959421

20 Library of Congress Country Studies (undated), Call Number DS706 .C489 1988;
http://lcweb2.loc.gov/frd/cs/cntoc.html#cn0058