Beruflich Dokumente
Kultur Dokumente
Contents:
Open mike roundtable by consortium members. Featured Practice Overview: Allen Hogg, Burke Interactive, Conducting Primary Business-to-Business Research Via the Internet Featured Practice Overview: Karl Feld, humanvoice, Inc. e-Interviewers: Adding the Interviewer Into Web-based Research. Featured Research Overview: Arvind Rangaswamy, Penn State. Rethinking Marketing Research for the Digital Environment.
Weve found so far that online market research is not much cheaper than using traditional methods, but we keep looking for good Internet research applications. (plastic packaging materials market research manager) Our business unit has not conducted online research yet, although other company units have. We plan to start with a value and loyalty study, comparing online to traditional methods. (plastics manufacturer eBusiness manager) Online focus groups look promising for researching some of our target markets, but we are concerned about the representativeness of online survey samples. (specialty chemical manufacturer market research manager) We use the Web for primary and secondary research of new markets. The Web allows more of a global information search, but the information it provides has holes we need to fill with traditional methods, such as phone calls. We also rely heavily on our salespeople for customer feedback. (specialty chemical company market research manager) Improve customer communication Were looking out for better techniques running customer surveys and discussions online. (information systems company product director)
Return to the Table of Contents
Market research on the Internet, growing an estimated 152% this year according to Inside Research, has exceeded forecasts over the past two years. Inside Research data are based on the spending of 29 research companies accounting for an estimated 90% of U.S. online market research. 83% is replacement research: Traditional or replacement research with data formerly collected by phone, mall, etc., but now replaced by online, plus new research methods only executable online. Concept screening, concept/product testing Advertising/brand tracking Customer satisfaction measurement Employee surveys Online focus groups 17% is Web-centric research: Web site measurement and evaluation. Profiling Web-site visitors
Web-site evaluation Banner ad testing Conducting Internet research can be Faster In 1998 study, Andrew Findlater, Reed Business Information received 160 of 400 required interviews in three hours Harris Interactive: 70%-80% of completions occur in the first 48 hours Generally, a majority of completions will occur within the first 48 hours Cautions: B2B research not quite as fast Do not stop interviewing just because quota has been met. Provide sampled respondents enough timeat least a weekto complete the survey before suspending data collection. Early respondents to email invitations might not be representative. Less expensive. The marginal costs of additional survey responses can be less than they are for mail or telephone. Incentives, however, can lessen or eliminate any cost advantages (Exhibit 1). Exhibit 1
89% with Internet access prefer to receive any survey via the Internet, according to Michael Foytik of DSS Research. White paper available at http://www.dssresearch.com/resources/bibliography.asp A study by Burke Interactive found greater likelihood of respondent participation in the same study in future research 33% of Web survey respondents saying they definitely would, 75% saying probably or definitely would. Just 18% of phone survey respondents saying they definitely would, 62% saying probably or definitely would. The survey, with limited open-ended questioning, took 12.4 minutes for average completion on the Web; 19.5 minutes by phone. [Comment by Arvind Rangaswamy of Penn State: There is some evidence that too many surveys on the Web are making the medium less productive, prompting some companies to return to phone surveys.] The trouble with traditional methods. Increasing non-response error. Phone refusal rates up, from 40% in 1988 to 46% in 1997, with no sign the trend is abating, according to The Council for Marketing and Opinion Research Mail response down. The Census Bureau assumes that about four out of 10 U.S. households that receive a 2000 census form wont fill it out, even though they are legally required to do so, American Demographics reported. Mall intercept completion rates down from 30 per day per mall in the 1970s to fewer than 5 in the 1990s, according to A.C. Nielsen. [Comment by Ada Nielsen of BP Amoco: Internet surveys are not as anonymous as mail surveys can be. Another problem is that Internet surveys work fine for business respondents using high-speed office connections, but are less satisfying if done at home over dial-up modems.] [Response by Allen Hogg: We have found there is little response difference between those who give their name online and those respondents who do not.]
Considered the Gold Standard for Internet respondents E-mail lists / databases Recruited from fewer sources Almost always recruited online, typically from a Web site People typically answer affirmatively that would be willing to be re-contacted for other research Contacted less frequently. Some panelists do not know they are in a database Less of a relationship. Survey cooperation rates can be much lower. A potentially good source for particularly targeted samples; e.g., diabetes sufferers Site-centric intercepts Typically used for Web-site evaluation, but could be used for any type of survey where Web-site visitors are the sample population Passive Recruiting (not recommended) Done primarily with the use of click me and interview me buttons. Drawback: respondent self-selection, akin to customer comment cards Benefit: less for the Webmaster to worry about Active recruiting (random or sequential) Done with behind-the-scenes software that is similar to Random Digit Dialing (RDD) techniques. Benefit: more representative ending sample Drawback: initially more work for the Webmaster Actively pulls aside every Nth visitor Cookies prevent respondents from being chosen more than once Able to sample visitors as they enter a site, exit a site, or both Response rates have run as high as 65% (avg. 25%) [Audience member comment: the technique described is more like interval sample than true randomization of dialing digits from a known database.] Customer lists / client databases / employee databases Merge and purge duplicate listings Make random selection from unduplicated list Send outbound e-mail invite with unique user ID embedded NEVER, NEVER, EVER include many addresses in a visible cc outbound e-mail Include URL (Uniform Resource Locator) that can be clicked on within email or cut-and-pasted into a browser. Many organizations are creating their own proprietary customer/industry interactive panels Ask about Internet usage during all research Ask respondents about their willingness to participate in future research Collect e-mail addresses of individuals interested in being included. Make access to topline information part of the incentive. Banner ads (not recommended) Self-selection bias.
Unknown population SPAM (not recommended!) Phone/mail recruitment to Web (not recommended) Weve tried, but response rates are low. [Acknowledging audience questions: We have not tried inexpensive postcard mailings. Nor have we tried recruiting at trade shows, but that might work.] [Comment by Karla Kuzawinski, independent research expert formerly at Xerox Corp.: We received a good response in B2B mailings, where customers clearly see the benefit of responding.] {Comment by Karl Feld, Humanvoice: We tried a mixed mode survey recruitment using mail and email, and found no difference. However, it was with a universe of loyal individuals, university alumni.] Mall intercepts, a consumer market option.
5. Provide specific instructions for any computer actions that may be necessary to answer a question 6. Put computer instructions with the question where the action is to be taken 7. Use caution when requiring respondents to provide an answer to each question before being allowed to proceed to the next question. Only do it if the answer determines question skip patterns. 8. Balance the use of a single scrollable questionnaire with a one question per page questionnaire 9. Be careful with questions with a large number of response alternatives. Be sure respondent can view all response options at the same time, on the same screen. 10. Let respondents know where they are in the questionnaire 11. Choose question types carefully and avoid problem questions. We found open-end question response better on the Web than in phone surveys. Offering a dont know option does not improve completion rates, we found. Additional best-practice suggestions: Allow sufficient time for responses; give all sampled individuals the opportunity to participate. Leave the response window open at least a week. Incentives and drawings can boost response rates, but be careful about violating lottery laws. Knowing the law is essential for international surveys Dont underestimate the importance of technical support; give respondents e-mail addresses and phone numbers to use if they encounter trouble. Try to use an FAQ link from the start of the survey to handle simple issues. Test the interview on as many browser/operating system/ platform combinations as possible. This is less critical a problem nowadays. Figure out how long it will take respondents using standard speed modems, and be honest about the length. Keep the interview short. Make user IDs and passwords good for one pass through survey only. Recognize that concept descriptions and graphic images should never be considered completely secure. Put your privacy policy statement on the questionnaire. Predicted respondent dropout rates. (Exhibits 2 through 5).
Exhibit 2
Exhibit 3
Exhibit 4
Exhibit 5
Pay attention to privacy concerns! Numerous organizations are policing the Internet regarding privacy Mail Abuse Prevention System (MAPS) LLC PrivacyPlace TRUSTe Online Privacy Alliance (OPA) Platform for Privacy Preferences (P3P) Electronic Frontier Foundation Better Business Bureau OnLine CPA WebTrust Governments are also interested U.S. Congress and the FTC European Union Directive on Data Protection Marketing research organizations have put forth ethical standards for Internet use; check their policies AMA (http://www.ama.org/about/ama/ethcode.asp) ESOMAR (http://www.esomar.nl/guidelines/internet_guidelines.htm) IMRO (Interactive Marketing Research Organization) http://www.imro.org/code.htm
CASRO (Council of American Survey Research Organizations) http://www.casro.org [Meeting attendee comments: incentives that have been tried frequent flyer miles computer printers charity donations (particularly important when government employees cannot accept gift) Amazon.com gift certificates Freestuff.com can draw freebie-hunting consumers to your survey, with surprising results. One chemical company doing a plastics study and giving away potholders to B2B respondents had to fulfill 60,000 for one days responses triggered by a Freestuff posting].
Is it valid research?
The Internet research peril: Biased data collected in vast amounts has absolutely no benefit to your organization! Sources of survey error: Sampling Error: sampling only a fraction of the population Coverage Error: excluding population segments from drawn sample Non-response Error: sampled individuals do not complete the questionnaire Measurement Error: a result of question wording or mode effects. Respondents might respond to scales differently on phone and Web surveys, for example. In many cases, all-Internet study designs introduce coverage error by excluding important parts of the population to be studied from participation. Although Internet penetration is increasing, many parts of the population including many companies customers and employees remain offline. Over-reliance on the large samples typically seen in Internet research can give researchers a false sense of confidence. It is quite possible that the one can precisely measure the wrong thing. Comparisons of survey methods: Comparing online, random digit dialing, and mail panel sample representativeness, Web respondents tend to be more educated, with larger incomes (Exhibit 6). Web respondents are more technically savvy, and use technology more. Demographic weighting probably cannot eliminate the discrepancy (Exhibit 7). Do not assume that an online survey response will be representative of online users. An online survey will likely not even generate a sample of Internet users similar to Internet users reached through traditional research methods (Exhibit 8).
Exhibit 6
Exhibit 7
Exhibit 8
Typically faster cycle times Typically a lower cost Enhanced ability to collect data across the world Switching modes can enhance the perceived importance of the survey and increase response rate Lower non-response error! Better enables Web reporting Potential drawbacks Potentially higher programming costs Greater coordination between sample control for the various methods is needed Greater attention to scaling and study design is needed Potentially higher Measurement Error: The result of inaccurate answers to questions that stem from poor question wording, poor interviewing, survey mode effects, and/or the answering behavior of the respondent (Dillman, op.cit.) Methodological differences (Exhibit 9). Phone seems to exhibit a recency effect while Web questionnaires, showing all response options, get more of a neutral response. Exhibit 9
Survey design implications. We recommend completely labeled and numbered scales (Exhibit 10).
Exhibit 10
Objectives: Maximize coverage Reduce non-response Minimize measurement differences No calibration of data needed Scales need to work well with both phone and Web administration Generally use completely labeled scales only when necessary (e.g., satisfaction) Avoid scales with seven or more scale points that are verbally labeled use end-point anchors instead. Construct surveys with eye toward unimode (Dillman, op.cit.) design (Exhibit 11). Note that the example does not permit a numerical scale. Create common question text Avoid the use of check all that apply type questions Avoid the use of drop-down boxes Avoid visual metaphors of scales Avoid the use of a single scrollable screen Use Web technology that can emulate CATI logic control (i.e., skip patterns should be passive) Email surveys might not be a good multi-mode choice
Exhibit 11
Near real-time allows for continuous data with a minor lag for automated data upload (i.e. results updated every day at midnight) Real-time allows for up to the minute changes in results (i.e. the last completed interview is available upon any request) [Comment by Karla Kuzawinski, independent research expert formerly at Xerox Corp.: Real-time reporting also provides an early warning system for survey problems that might arise.]
Considered, precise, unembarrassed comments. A more natural setting for some (e.g., software developers). Equal air time available for all participants. Circumvent problems caused by group dominators. Can incorporate powerful qual/quant hybrids Can gauge reactions to multimedia stimuli on World Wide Web. Frequently used for Web-site evaluation research. Immediate electronic transcripts are natural by-product. Every comment can be associated with a writer and the time when it was written. Verbatims can be easily inserted into electronic reports, posted on Intranet sites. Preparation Screen for chat room experience Have people test out access ahead of time Work to avoid losing respondents due to distraction. Remind them that incentives are available only to those who stick with it. Create detailed discussion guide from which you can cut-and-paste. Questions to consider when choosing software and suppliers. Will the focus group site be secure? Are participants given user names and passwords? Will participants need to download any software? Is it possible to establish private channels? Can you send private messages to the moderator? Can the moderator send private messages to individual participants? Can you throw out problem participants? As for online research partners: Are they good researchers? Do they provide research solutions, or want you to buy Internet research? Online experience is necessary, but not sufficient. Can they provide data from multiple streams (e.g., Web, phone, IVR, etc.)? Are they informed about the differences between methods and can they help you expect to see certain effects? Can they add value to the data through analysis? Do they respect privacy issues and will they support industry standards? How do they provide sample (intercept, recruit, panel, etc.)? Are they versed in the issues of online sampling? Where will the data reside, and is it secure? What are their interviewing capabilities? Similar to CATI, or limited to simple questions? Is their interviewing platform stable (24/7)? Can they provide research results online? Consider the Internet buzzwords of 2000 Broadband
Mainly now ISDN (integrated services digital network), DSL (digital subscriber line), and cable modem Billions now being spent on broadband satellite systems 11% of U.S. households with Internet access now have such broadband connections (Statistical Research Inc.) Enables online testing of commercials and other multimedia products, but Coverage error concerns of general online research can be magnified Wireless By next year, 1.3% of U.S. residentsand 8% of Western Europeanswill be using wireless Internet services (Forrester Research) Numerous companies are working on survey technology for personal handheld devices Potential for survey respondents to be using very small screens reinforces the notion that questionnaires should be designed simply
Return to the Table of Contents
Text chat with interviewer, a live person watching the respondents progress through questionnaire. The interviewer might send encouraging messages as appropriate. Click option for respondent to ask interviewer to call on telephone, for help, etc.. (The respondent can designate when the interviewer should call.) The interviewers picture to humanize the contact. Demonstration and white paper available at www.surveyguardian.com
Known technical barriers higher with e-interviewer because interviewers could ask respondents. Self-administered Web questionnaire required twice as many respondents going in. Technical problems Retention and completed response problems. Larger ineligible respondent count with self-administered survey increases sample requirements. Exhibit 12
Exhibit 13
e-interviews solve causes of mid-termination No rapport with interviewer Presence of open-end questions Respondent technical difficulties No opportunity to request question clarification or verification of sponsor. Current humanvoice system provides phone link when respondent requests a call. Future generation of software will include a voice IP connection. No option for respondent to change interview mode at will (Exhibit 14). Exhibit 14
Solving some knotty Internet challenges. Reducing mid-terms decreases sample panel fatigue and self-selection bias. Click to callback and instant messaging allow resolution of technical problems or interview mode change. Increase by 30% the number of respondents 55+ years who complete surveys. Increase open-end response length Important to B2B surveys. Consistent with previous research, findings indicate that e-interview open-ended response exceeds phone surveys.
Fewer terminations on open-end questions when interviewer present. Average word length of open-end responses and follow-up probes is longer (Exhibit 15). Exhibit 15
Screen out ineligible and professional respondents. Frequency of professional respondents in the U.S. is on the rise Self-administered surveys have always suffered from ineligible respondents Self-selection rates are high in sample panels with high mid-termination rates Site http://www.money4surveys.com brings professionals to your site who want the money. Ineligible respondents caught by version. (Security required unique PIN and year of birth.) CATI: n=300 & 5 ineligibles e-Interviewer: n=117 & 2 ineligibles Self-Administered: n=122 & ?????????
It is inexpensive
For a survey featuring n=600 50 questions Listed, 80% incidence sample Re-invite after initial sample send Using SurveyGuardian 24/7 these are comparative costs: CATI: $11,356, including the high cost of callbacks. e-Interviewer: $8,414.
Self-administered: $4,850. The fixed cost of programming can drive up the cost per response for small surveys. Costs will be different, probably higher, for B2B surveys.
Additional benefits
Open-ended questioning capability allows customer service options, such as immediate response to a customer complaint. Interviewer-collected data is cleaner in the database.
Return to the Table of Contents
Key points
Traditional research migrating to online environment is analogous to the early years of television. TV started with radio formats exported to the new medium, until it developed its own unique formats. Pushing the online research envelope is the real golden goose promised by the medium. Myths about the online marketing environment are being challenged. Research finds: The more information buyers receive online, the less important the brand name as a cue to quality. Its total nonsense that online markets increase buyer price sensitivity. Buyers become more value sensitive. Price is considered a correlate to value, but the additional product information available online allows buyers to weigh other value factors as well. They dont necessarily buy the cheapest offering. About 60% of buyers use the Web for convenience; they do not use all the buying information available to them online. About 40% use Web information to optimize choices.
There does not appear to be a difference in customer satisfaction online or offline, suggest studies by Marriott. But online customers tend to be more loyal because, it seems, its easier to remain loyal online if youre satisfied.
Exhibit 16
Web site evaluation surveys Content and Structure (Examples of items) Graphics Visual Attractiveness Selling Messages Links Chat rooms Registration Forms Audio Good sources for Web site evaluation: http://www.iconocast.com/ and http://www.emarketer.com/ Ease of navigation (e.g., Site search engine) Experience during visit How much did respondents enjoy their visit? Did visitors feel confused while using the site? Were visitors frustrated in any way with their experience? Did visitors find their visit exciting or boring? Did the site meet, exceed or fall short of visitors' expectations? What was visitors' overall level of satisfaction visiting the site? Likelihood of repeat visit
Polls apart?
Non-representative samples
Self-selection bias. Respondents are heavier users of computers, Internet and email than non-respondents Matching a sample to population on observable characteristics will not make it representative (e.g., propensity weighted scores wont work!) There are unobserved heterogeneities, but how can we observe what we cannot see? We must use methods to differentiate Web users from non-users. Why do some use the Web and others do not? Low response rates (e.g., banner clicks average around 0.30%) Problems of respondent authenticity Difficulties associated with incentives Difficulty in gauging response accuracy
www.modalis.com: Web-based surveys, e-mail surveys. Respondents recruited on the web. www.comscore.com: Generates company-specific panels and monitors their web behavior. www.greenfield.com: 2.2 million panelists who have volunteered to be members. Participates in drawing to win cash prizes. www.harrisinteractive.com: 7 million panelists worldwide.
Select samples from panels Pre-qualify and profile respondents Put banners/links to survey at popular web sites Offer incentives for participation. Even $2 can boost response. Focus on products of general usage (e.g., supplies) and segments that match the Internet population.
Reduce travel costs Useful for discussing sensitive issues (requiring some anonymity) Quick turnaround (e.g., transcripts and keywords) Challenges Changes the dynamics of the communication: typing vs. voice. Handling emotive issues Some innovations Focus groups via videoconferencing Focus groups via chat windows
More innovation
Test market experiments Electronic shelf labels Computer-simulated test markets: a stable technology now that could have value for B2B, such as simulated trade shows. Promotional kiosks Customer tracking: e.g., point-of-sale linked to infrared sensors, now experimental.
Status code - the code for the resulting success or failure of the request Transfer volume - the number of bytes sent to the reader's browser as a result of the request. Sample log, Exhibit 17 Exhibit 17
Statistics from common log format Number of requests Number/percentage of successful/failed requests Number/percentage of cached requests Top pages or files (most requested documents) Number of page-transfers by day Top downloaded files by type (all files) Top submitted forms and scripts Bottom pages or files Top pages by directory Top directories accessed Average number of requests per week Average number of requests per day Total bytes transferred Average bytes transferred by day Average bytes transferred by hour of day Average number of hits on weekdays/weekends Most/least active day of the week (and number of hits) Most/least active day ever (and number of hits) Activity level by day of week/hour of day
Enhancements to the common log format Difficult to link information across log files Combined Log Format (Combines Transfer log, Referrer log, and Agent log). Difficult to identify unique visits Cookies (Stored in browser with expiration time) CGI Session ID (appended to URL) User-registration (Append ID information from any of these methods to log files) Statistics from enhanced log format Number of visits Average number of requests (and page views) per visit Average duration of a visit Sequence of user activities at the site Average number of visits per day or week Number of visits by hour of the day Visits from organizations (most active organizations) Visits by organization type (root domain) Visits from countries (most active countries) Top visit entry pages Top-page durations Top exit pages Average number of users on weekdays/weekends Visit level by day of week/hour of day Top U.S. geographic regions Percentage of visits from inside/outside the U.S. Top cities Top referring organizations Top referring URLs Top browsers Top user operating systems But statistics only tell you what people do, not why they do it. To derive marketing insights from Web site statistics, you need combined market data sources: e.g., Exhibit 18.
Exhibit 18
Access to volumes of secondary research Potential for inexpensive, instantaneous, interactive, and global communication with customers More realistic marketing stimuli and decision contexts Ability to dynamically change marketing programs and measure consumer response Individual-level data on search and choice
Establish data sharing learning communities for benchmarking. Link everyone in an iterative learning process, for example. Take advantage of the developing network of respondents. Do online experiments Develop methods and models for deriving insights from large data sets Create customized marketing research bots Real-time research and analysis: e.g. www.valueharvest.com indicates how data and analysis can be stored and run on different servers (Exhibit 19). Exhibit 19
Co-design products with customers. Texas Instruments designed a calculatorits most successful modelfor the educational market, with 30,000 educators responding to TIs invitations to comment. Factors going in Complex buy cycle Long adoption/legal issues Mixed price sensitivity Need support/curriculum Vendor credibility is key Active community of customers
Insights gleaned The market owns the product! Therefore, the product owns the market. Reduces adoption times Web can concentrate influential audience Need responsive design team that can deliver Need a culture of experimentation Netscape stumbled into customer co-design with its version 3.0, because it had many engineers online. Establish data-sharing learning communities for benchmarking. Create virtual learning communities. Example: Companies pool their Web use information in the benchmarking survey of Penn States eBusiness Research Center (www.ebrc.psu.edu/benchmark) A model for sharing information, the survey software gives respondents an instantaneous report of all respondents collective input, as illustrated in Exhibit 20. Reports preserve respondent anonymity. Exhibit 20
Q&A
Q: How do you develop or find good B2B email lists?
A: Make your Web site the dominant information source in your market. Attract audiences by becoming the destination site.
Return to the Table of Contents