Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
29 C H A P T E R 4 Survey invitations were sent to all addresses obtained from DirectMail within the defined study areas. Respondents were assigned one of four versions that presented a different series of six roadway images. Invitations for the first wave of the survey were sent in Summer 2016, with printed copies mailed in the fall to all those who had not responded. Invitations for the second- wave survey were sent in Spring (or, in the case of Chattanooga and Birmingham, Fall) 2018 to all those who responded to the first wave. Wave 1 Respondents In total, 1,223 respondents completed the survey: 178 online and 1,045 on paper. Responses were distributed by area as shown in Table 4.1. Although each area received lower than the desired 10% response rate, this sample is large enough to provide statistically meaningful results from the survey from all areas. The response in Talladega is enough for a control area, although segmentation by demographics or other variables was limited. As discussed previously, four different survey versions were used to limit the number of images that any one respondent saw. The four versions were evenly divided among the six areas. As shown in Table 4.2, the responses were fairly evenly distributed as well. Two respondents dropped out of the online survey before being assigned a version. The sociodemographic section of the survey was rather extensive, though a summary of several key sociodemographics is presented in Table 4.3. (A detailed presentation of demographics can be found in Appendix D.) As is typical for self-administered surveys of the general population, the sample and the population are noticeably different on several variables. The small P-values for the chi-square goodness-of-fit tests on the presented demographics indicate a deviance of the sample from the population. First, the respondents tended to be wealthier than the study- area populations. Second, inspection of the data indicates that one-person households were underrepresented in the sample, while two-person households were overrepresented. Respon- dents also tended to be older than the population of the combined study areas, which is common in surveys like this one, as older individuals are more likely to have more time to respond. The average age of the survey respondents was 52 years old. Finally, a greater proportion of white/ Caucasian residents responded to the survey than those of other races and ethnicities. The non-representativeness of the sample somewhat limits how accurately the data can describe the overall population. However, when applied in an explanatory sense, as with the regression models in this study, the âpotential defectâ presented by such a sample âis less significant than it would be in descriptive researchâ (Babbie 2010). Nonresponse bias limits the Description of Survey Respondents
30 Bicyclist Facility Preferences and Effects on Increasing Bicycle Trips Area Households Contacted Responses Response Rate Anniston 4,348 198 4.6% Opelika 3,362 185 5.5% Chattanooga 4,400 239 5.4% Talladega 3,305 93 2.8% Northport 3,708 234 6.3% Birmingham 4,294 274 6.4% Total 23,417 1,223 5.2% Note: Anniston, Opelika, and Chattanooga are treatment locations with planned bicycle facility improvements. Table 4.1. Raw survey response count by area. Version Number Responses Percent of Total 1 305 25.0% 2 317 26.0% 3 314 25.7% 4 285 23.3% Table 4.2. Raw survey response count by version (N = 1,221). Responses % Respondents Answering Questiona % Population from ACS Household Income (N=1,146; Chi-square goodness of fit P<0.001) $15,000 or less 179 18% 29% $15,001â$30,000 140 14% 23% $30,001â$50,000 151 15% 19% $50,001â$75,000 177 18% 14% $75,001â$100,000 118 12% 6.4% $100,001â$125,000 69 7.1% 4.2% More than $125,000 142 15% 5.6% Household Size (N=1,130; Chi-square goodness of fit P<0.001) 1 person 422 37% 40% 2 people 473 42% 31% 3 people 97 8.6% 14% 4 people 77 6.8% 9.7% 5+ people 61 5.4% 6.1% Respondent Age (N=1,113; Chi-square goodness of fit P<0.001) 18â34 204 18% 40% 35â49 214 19% 21% 50â64 378 34% 23% 65+ 317 28% 16% Respondent Race/Ethnicityb (N=1,145; Chi-square goodness of fit P<0.001) Black/African American 312 27% 52% White/Caucasian 771 67% 43% Hispanic/Latino 9 0.8% 4.6% Other 67 5.9% 4.6% a. Excluding those who specified âPrefer not to answer.â b. Respondents were allowed to mark more than one (percentages may exceed 100%). Table 4.3. Sociodemographics for pooled sample and American Community Survey population.
Description of Survey Respondents 31 ability of the study to describe the preferences of the population as a whole, but not necessarily to explain the relationships between variables. Including sociodemographics in models is a time-honored way of controlling for nonresponse bias. Vehicle and bicycle ownership are presented in Table 4.4. Most households owned one or two vehicles, although a modest portion did not own a vehicle. More than half of the house- holds did not own a bike. Nevertheless (table not presented), about 11% of respondents reported bicycling for utilitarian purposes at least monthly. Additionally, nearly 20% reported cycling for recreation to some degree. However, only 1% of respondents reported daily utili- tarian cycling. This discrepancy, between the number of daily utilitarian cyclists and both the number of respondents who cycle infrequently and the number who have access to a bike in their household, indicates a sizable persuadable portion of the sample is capable of biking but does not bike regularly. Data Cleaning A general screening and more in-depth review for missing data were used. Unfinished surveys and those with a low portion of questions answered were removed entirely from the raw data- base. An additional assessment was undertaken section by section, using commonly accepted methods to fill in small amounts of missing data and excluding cases with an unacceptable amount of missing data. Cases were evaluated for inclusion or imputation on different comple- tion criteria for each section, as follows: â¢ Section A (Attitudes): Cases with more than five missing items were deleted. Otherwise, missing items were imputed using expectation maximization. â¢ Section B (Technology usage): Uncleaned. â¢ Section C (Home): Uncleaned. â¢ Section D (Daily travel): Uncleaned. â¢ Section E (Bicycle experience): For key dependent variables and segmentation variables, all missing responses were excluded from the respective models. â¢ Section F (Demographics): Where available, responses with small amounts of missing sociodemographic data were supplemented with information from our targeted marketing database. After cleaning, the raw database was consolidated into a working database of 1,178 respon- dents. Each person responded to six different images, so there were up to 7,068 image responses for each of the four questions (comfort, safety, willingness to try, and frequency), though cases were excluded from the respective models of those variables owing to item nonresponse. Number of Vehicles Responses % of Respondents Number of Bikes Responses % of Respondents 0 124 11% 0 618 53% 1 399 34% 1 240 21% 2 413 36% 2 163 14% 3 141 12% 3 69 6.0% 4 52 4.5% 4 33 2.8% 5 16 1.4% 5 13 1.1% 6 8 0.7% 6 15 1.3% 7+ 6 0.5% 7+ 8 0.7% Table 4.4. Number of vehicles and bikes owned by survey respondents (N = 1,159).
32 Bicyclist Facility Preferences and Effects on Increasing Bicycle Trips Wave 2 Respondents Invitations for the second-wave survey were sent to all those who responded to the first-wave invitation. The second-wave survey had just under a 50% response rate, as shown in Table 4.5. Since survey invitations were sent by mail, it was inevitable that some follow-up responses would be filled out by a different respondent than the first survey. These potential respondents were identified and excluded from any analysis of change, but are still included in Table 4.5. For full details of the demographics of matched respondents, see Appendix D. A summary of demographics for respondents to both waves is presented in Table 4.6. Respon- dents in both waves appear similar, but inspection of the statistics reveals some discrepancies. Area Households Invited Initial Responses Initial Rate Follow-up Responses Follow-up Rate Anniston 4,348 198 4.6% 98 49.4% Opelika 3,362 185 5.5% 103 55.6% Chattanooga 4,400 239 5.4% 85 35.6% Talladega 3,305 93 2.8% 47 50.5% Northport 3,708 234 6.3% 145 62.0% Birmingham 4,294 274 6.4% 105 38.3% Total 23,417 1,223 5.2% 583 47.7% Note: Anniston, Opelika, and Chattanooga are treatment locations with planned bicycle facility improvements. Table 4.5. Invitations and responses for the first and second survey waves. % Respondents Second Wavea % Respondents First Wavea % Population from ACS Household Income (Wave 1 N=1,146; Wave 2 N=457) $15,000 or less 13% 18% 29% $15,001â$30,000 14% 14% 23% $30,001â$50,000 16% 15% 19% $50,001â$75,000 18% 18% 14% $75,001â$100,000 12% 12% 6.4% $100,001â$125,000 8.5% 7.1% 4.2% More than $125,000 18% 15% 5.6% Household Size (Wave 1 N=1,130; Wave 2 N=542) 1 person 37% 37% 40% 2 people 43% 42% 31% 3 people 10% 8.6% 14% 4 people 6.1% 6.8% 9.7% 5+ people 3.3% 5.4% 6.1% Respondent Age (Wave 1 N=1,113; Wave 2 N=565) 18â34 11% 18% 40% 35â49 18% 19% 21% 50â64 29% 34% 23% 65+ 42% 28% 16% Respondent Race/Ethnicityb (Wave 1 N=1,145; Wave 2 N=574) Black/African American 22% 27% 52% White/Caucasian 73% 67% 43% Hispanic/Latino 3.3% 0.8% 4.6% Other 4.3% 5.9% 4.6% a. Excluding those who specified âPrefer not to answer.â b. Respondents were allowed to mark more than one (percentages may exceed 100%). Table 4.6. Sociodemographics for first and second waves and American Community Survey population.
Description of Survey Respondents 33 A smaller portion of respondents in the lowest income group responded to Wave 2 than to Wave 1, indicating continuation of a response bias against this lowest income group. The distribution of response rates among households of different sizes was somewhat stable between waves. Respondents in the second wave tended to be older than those in the first wave, which may be partly explainable by respondents aging 1 to 2 years between waves. Yet the average age of the survey respondents was 56 years old, which implies that the aggregate change of the age distribu- tion is larger than just the effect of aging. Race and ethnicity breakdowns were similar between waves, though with some drastic movement among Hispanics/Latinos and other races and a modest shifting of a larger portion of Caucasian respondents than African Americans. Vehicle and bicycle ownership are presented in Table 4.7. Distributions of vehicle ownership did not appear to change drastically between waves, though a surprisingly higher percentage of second-wave respondents reported not owning a bike than in the first wave. Table 4.7. Number of vehicles and bikes owned by respondents for first and second waves. Number of Vehicles % Respondents Second Wave (N=568) % Respondents First Wave (N=1,159) Number of Bikes % Respondents Second Wave (N=567) % Respondents First Wave (N=1,159) 0 11% 11% 0 57% 53% 1 32% 34% 1 19% 21% 2 38% 36% 2 12% 14% 3 11% 12% 3 4.4% 6.0% 4 4.9% 4.5% 4 4.2% 2.8% 5+ 3.0% 2.6% 5+ 3.2% 3.1%