CDT Data Analysis: where is the bubble?
tl;dr: I’ve looked at hiker comments across the CDT to assess hiker trends. About 1000 hikers start on the CDT every year. Only around half of all hikers complete the trail in a given year, and there’s around 100-300 flip-floppers and skippers across sections. I made a little web widget (link at bottom) to play with that estimates where the bubble is on any given day. I also have a full, very long write-up that covers more analysis and information about the data and methods on my website (also linked at the bottom). This post is an overall summary of key bits of analysis, which I think folks other than myself may also find interesting!
Data Summary
I went ahead and gathered hiker comment data over time. This gives me a log of comments recorded based on trail miles, dates, and unique hiker identifiers. This data set can allow for some interesting analysis that lets you pool statistical relationships together, to estimate different things about the trail!
Where’s the Bubble?
I used a variety of methods to estimate the bubble. This included 1) looking at the pace of hiking, by sampling hikers that made comments on consecutive days, and calculating the distance in miles between those consecutive comments, and 2) looking at the date distribution for comment density as you progress up the trail (different comment dates create the shape of hiker distribution for a given mile). This table is a rough estimate for the expected median date that the bubble might be at any mile, in an average year on the CDT, based on the northbound median comment date.
| Mile | Expected Median Date for Bubble |
|---|---|
| 40 | 21-Apr |
| 540 | 20-May |
| 1,040 | 26-Jun |
| 1,540 | 13-Jul |
| 2,040 | 4-Aug |
| 2,540 | 22-Aug |
Estimating where hikers are, over time, on the CDT
How does snow impact the bubble
I used NOAA SNODAS data, and pooled different Snow Water Equivalent (SWE) estimates on April 1st to provide an estimated SWE prediction effect relative to the median SWE for each section between 2005–2026. I incrporated the SWE ratio value to see if this would improve the model's ability to predict a median data for each relevant section compared to the model without the SWE data. I found that snow data had an effect for a few sections of the CDT. THe Table below shows how much snowpack impacts the median for the 4 relevant sections of trail where it was found to meaningfully improve the model. The results suggest that for a big snow year, the hiker bubble seems to be delayed by around 10-or-so days relative to an "average" year.
| Stretch | Median-of-other-years error | With snowpack | Days of delay estimate per doubling of snowpack |
|---|---|---|---|
| CDT, mile 1,000 (9 hiking years) | 2.7 | 2.1 (better) | +4.4 |
| CDT, mile 1,100 (8 hiking years) | 6.8 | 5.3 (better) | +10.0 |
| CDT, mile 1,200 (8 hiking years) | 5.9 | 4.4 (better) | +9.5 |
| CDT, mile 1,300 (8 hiking years) | 4.4 | 4.0 (better) | +6.2 |
How many folks skip and flip, and where to?
Because I can estimate the standard hiking pace for hikers, I can look at the median distribution in one part of the trail, and estimate what the distribution should look like further up the trail by using the standard hiking pace. Additionally, as a check, I can also look at how the median distribution of comments moves from one section of trail to another, and look at the time between those two medians, and the number of miles, to estimate how quickly the bubble had to move to exhibit that sort of behavior. For areas of trail where the hiker distribution seems to deviate from the statistically-likely appearance, we have good reason to believe skipping or flipping may be occurring. The table below estimates the number of hikers that flip or skip to a different section of the trail based on conditions in prior sections of trail. The comment data reflects high likelihood of skipping and flipping in Colorado, and also around the Wind River Range.
| Stretch | Median date at the near mile (A) | Median date expected at the far mile (B) | Median date actually seen at B | Days ahead of the expected bubble | Extra people ahead of the crowd (area) | Share needed to cross the median to explain the shift | Rough number of flip-floppers per hiking year |
|---|---|---|---|---|---|---|---|
| CDT, Pagosa Springs to Salida (miles 800 to 1,000) | Jun 14 | Jun 27 | Jun 22 | 5 | 19% | 18% | 50 to 100 |
| CDT, Grand Lake to Rawlins (miles 1,300 to 1,500) | Jul 9 | Jul 19 | Jul 14 | 5 | 13% | 17% | 20 to 90 |
| CDT, Pinedale to West Yellowstone (miles 1,750 to 1,950) | Jul 29 | Aug 7 | Aug 2 | 5 | 16% | 19% | 30 to 80 |
How many people injure off/leave the trail?
To estimate the number of people dropping from trail, I look at the commenters and the frequency of their commenting behavior. Significant changes in their commenting behavior over time (goes from commenting a lot to having no more comments), is a good indicator that something happened (e.g. they injured off trail, or had a serious illness).
To estimate the highest likely rate of attrition, I took everyone who commented near the start of the trail. This population is established by creating a cohort for the northbound-looking people with a comment in the first 60 miles. If a person has a comment at or beyond a certain mile, that person is still considered to be “on trail” and hiking. Someone who never comments again is considered to have left the trail. As such, this bound overestimates attrition.
To estimate a conservative rate of attrition, I looked at a cohort of frequent commenters. This cohort consists of people with at least six comment-days and a typical gap of four days or less between comments across the trail. If one of these “frequent commenters” goes quiet, it implies something changed. This is likely to be indicative of an actual attrition event. Because hikers in this cohort had to be hiking long enough to write six times with gaps of 4 days or less between those comments, this is a conservative estimate of attrition. As such, this metric underestimates the number of people leaving the trail. The table below estimates the likely range of people staying on trail relative to those that initially started.
| Trail, mile | Survivor Rate (people that ever commented) | Survivor Rate (frequent commenters) |
|---|---|---|
| CDT, mile 300 | 85% | 96% |
| CDT, mile 500 | 80% | 93% |
| CDT, mile 1,000 | 68% | 81% |
| CDT, mile 1,500 | 57% | 68% |
| CDT, mile 2,000 | 47% | 58% |
| CDT, mile 2,400 | 45% | 56% |
Probable upper and lower bounds for CDT hiker attrition
There's a lot more analysis and discussion available on my very, very long blog post about it. The web tool is much easier to use to give you a quick snapshot of "where is the bubble?" And to estimate how many people might be ahead of or behind you at any given point on trail.
^(Generative AI Disclosure: the front-end tool I created to explore bubble trends was built largely with the help of generative AI. The underlying data is actual hiking data that were not generated. Any errors or inaccuracies are primarily the result of data quality and suitability of the analysis methods used. The methodology is repeatable and consistent, and several analytical approaches were conducted to assess statistical robustness. All writing was done by me. GenAI was used to help with formatting, graphing, and organizing of the main blog post.)