Internet Access

1
Internet Access

• Is everyone online yet and can we survey them
  there?

TRB Planning Applications 2009 Session 9A Travel
Survey Methods and Analysis II
Colin Smith Greg Spitz Resource Systems Group,
Inc. Thanks to Mark Fowler Tyson Seely of
Resource Systems Group Prepared for TRB
Planning Applications Conference May 18, 2009
Greg Spitz Director of Transportation Market
Research RSG, Inc
2
Research Question

• For transportation surveys
• Can we obtain a valid sample if we only recruit
  and survey respondents who have internet access?
• What are the populations we are interested in
  sampling?
• Toll road customers
• Transit riders
• Potential users of a new or changed road or
  transit system
• Geographical areas of travelers

3
Research Question

• What do we mean by recruit?
• Contacting potential respondents by mail,
  directly on facilities, at activity locations,
  through customer lists, or through paid sample
  providers
• Why would we only want to recruit respondents who
  have the ability to respond online?
• Online surveying is generally less expensive
  than other methods and can be used to administer
  very complex survey instruments

4
Research Question

• What is the problem we are concerned about?
• Coverage error
• Dillman describes coverage error as, the result
  of not allowing all members of the survey
  population to have an equal or known chance of
  being sampled for participation in the survey
• Do we introduce coverage error if we only
  recruit respondents with internet access?

5
Coverage Error Telephone Surveys

• Rapid growth in the prevalence of cell phones
  raises concerns about undercoverage in RDD
  telephone samples (Tucker et al, 2007)
• Wireless-only households made up 15 of US
  households in 2007, significantly higher in 2009
• Undercoverage is clearest for low income young
  adults proportion of wireless-only households
  was 32 in 2006.

6
Coverage Error Telephone Surveys

• Rate of increase in wireless-only households is
  rising rapidly in the recession
• Additional15 of households have both landlines
  and cell phones but take few or no calls on their
  landlines (e.g., fax/internet line)
• Attempts are being made to supplement RDD
  samples with cell phone samplesbut cell phones
  are not geographically based and are more costly
  and produces relatively low rates of
  participation (Link et al, 2007)

7
Coverage error Internet Surveys
Source Pew Internet and American Life Project
8
Coverage error Internet Surveys
Those over 65, with low income and with lower
levels of educational attainment have lower rates
of internet access.
Source Pew Internet and American Life Project
9
Case Study 1 Toll Road Market Research
Computer-based stated preference surveys designed
to estimate the values of travel time savings of
potential users of road-pricing projects
10
Project Locations
Anchorage, AK
San Francisco, CA
Denver, CO
Atlanta, GA
Jackson, MS
Orlando, FL
Broward County, FL
11
Recruitment Approach

• Recruitment was carried out in several ways for
  these surveys the samples can be grouped into
  four categories

The analysis presented today is based on the data
collected at activity locations (e.g. shopping
malls) where respondents are intercepted at random
12
Internet Access Questions
13
Internet Access Questions
14
Internet Access
15
Internet Access
16
Demographics With and Without Internet Access
17
Demographics With and Without Internet Access
18
Demographics With and Without Internet Access
19
Case Study 1 Conclusions

• Toll road studies have higher proportion of
  internet access
• This is likely due to income effects of vehicle
  ownership
• Internet access sample is higher income and owns
  more vehicles
• The sample with internet access is not
  significantly different along income, age, and
  vehicle ownership dimensions than the full sample
• Collecting data only from an internet access
  sample would not lead to significant coverage
  error, but does undercover a small and different
  part of the population

20
Case Study 2 Transit On Board OD Studies
System wide OD study of all CTAs bus and subway
lines
Chicago, IL
New York City, NY
OD study of Metro North Railroads three
east-of-Hudson commuter rail lines
21
Recruitment Approach

• Transit origin-destination surveys
• Large customer sample
• Travel patterns obtained
• Onboard recruitment
• Large samples 92,000 responses in New York and
  34,000 in Chicago
• New York all inbound Metro North trains
  surveyed
• Chicago a sample of runs of every CTA bus route
  and subway line

22
Contact Information Questions
23
Contact Information Questions
24
Coverage Error Phone Respondents vs. Email
Respondents

• Telephone respondents are not landline-only
• Research question is the email sample
  representative of the overall sample, or would
  surveying by email lead to coverage error?

25
Coverage Error Phone Sample v Email Sample
26
Metro North Railroad Phone and Email Samples
27
Metro North Railroad Phone and Email Samples
28
CTA Phone and Email Samples
29
CTA Phone and Email Samples
30
Case Study 2 Conclusions

• Large samples mean high confidence that overall
  sample is representative
• Very different demographic profiles across the 2
  studies
• In both studies, email and phone samples are
  representative of the overall sample
• Collecting an email or phone contact list for
  future market research from surveyed transit
  riders does not lead to significant coverage
  error

31
Research Conclusion

• Do we introduce coverage error if we only recruit
  respondents with internet access?
• Two cases
• Potential road users intercepted at activity
  locations such as shopping malls
• Transit riders surveyed on board the transit
  vehicle
• In both cases, the sample of survey respondents
  with internet was very close to the overall
  sample, though weighting of income and age should
  be conducted
• Therefore, we would not introduce significant
  coverage error if we only surveyed those with
  internet access, and this can be mitigated
  through weighting to population proportion.

32
Final Thoughts

• No survey method completely eliminates coverage
  error
• Anecdotally, web is perceived to have the
  highest coverage error in our industry
• Reality is that the web is as good or better
  than commonly used RDD and other survey methods
  to represent transportation study populations
• Broad based geographic studies are the most
  difficult to sample. The best way to sample such
  studies is to use addressed-based sample and not
  RDD or purchased sample
• Many broad based studies still conducted using
  RDD and this creates serious coverage error
  issues