Cycle tour through multivariate statistics: Part 1 Data Preparation

Last updated on August 30th, 2019 at 07:00 am

As part of my statistics course this semester, I did a survey of cycle tourists and used a variety of multivariate statistical techniques to analyze the resulting data. First off, I’d like to thank those of you who responded to the survey, special thanks to warmshowers.org, crazyguyonabike.com, travellingtwo.com and goingeast.ca, for sharing the links to the survey. I was overwhelmed with the response, with the initial count indicated 222 responses. In this series of posts, I’ll share my narrative of how I analyzed the data using the various statistical techniques. If you don’t care about how I did it, you just want to see the results, here is the link (will be active once part 8 is posted). If you’d like a copy of my detailed report – leave me a comment.

Data Preparation

In the cycle tour dataset, the initial data indicated 222 responses; however, there were several responses that had mostly blank data. After removing these blank responses, I was left with 207 responses. In addition, in order to reduce excessive outliers (that is, extreme values that break the statistics), I had to remove rows that did not fit within the bounds of the dataset. I defined the requirements as: total duration of at least 7 days, minimum average daily distance of 20km/day (based upon TotalDist/TotalDays), and a maximum average daily distance of 150km/day. I justify the minimum duration as part of the definition of a cycle tour, those traveling less than 20 km/day likely spend most of their time not cycling, and those averaging more than 150 km/day are likely randonneuring rather than cycle touring. In addition, anyone not answering the TotalDist or TotalDays questions had to be removed. That left me with 170 responses. Next I needed to test for outliers. I used SPSS to generate a list of possible outliers. I checked only the unbounded variables (Age, Panniers, Budget, TotalDays, TotalDist, and Variety). I then inspected the Extreme Values table. The output identified the following outliers:

• Budget had two extreme values ($200 and $300). Since this value was significantly higher than the rest, and there was no clear reason why and I could not justify changing the result, I deleted these two records.
• Panniers had one outlier, a value of 8. Given the lack of precision in the calculation of panniers, I chose to modify this value to be the highest less extreme value, 7.
• TotalDays had three extreme values (1598, 1207, and 1100). Since each of these tours indicates a trip of duration longer than 500 days, but would have included a trip of 500 days, I chose to modify these values to be the highest less extreme values of 503, 502, and 501 days respectively. In doing so, I also adjusted the TotalDist value and variety values by the same ratio.

After making the adjustments, I re-ran the explore statistic to validate the absence of outliers. The dataset had 168 respondents. The dataset was cleaned up and coded into SPSS with the following variables: