High vs. Low Tire Pressures: Which Is Faster?

I’m not sure why you’re hung up on this idea of less data. Recall the purpose of a glm wrt data generating processes and inference. We specify and diagnose a model such that it is logically equivalent to the true data generating process. When you have the correct model specified, you can simulate the data as much as you want or continue to collect runs and the results will be robust. This is a major difference between generative and discriminative models - DoE is built on the former. 

Either way, in simpler terms, you’ve gone thru a lot of work to create an analysis that only confirms what riders already believe. The analysis, while nice and confirmatory, adds nothing to the current body of knowledge regarding tire pressure on mtbs. So re: cost/benefit analysis, this experiment has cost but no benefit. 

… I know this is a weird format for technical feedback and I appreciate your candor. But I’m notanengineer 😂, and although I would like to encourage you to keep experimenting, this kind of experimentation doesn’t fly in industries that rely on DoE for major decision making. 

I'm around the same weight, but heavier.  6'1, 225.  I typically run 23 on the front, and 27 on the rear.  That's my typical xc/all mountain pressure setting.  I don't really ride enough DH or Park these days to know what works best for that, but typically I just stick to the same pressure all the time.

I agree with you about front tire traction completely.  I've been saying the same thing for almost 20 years.  My front traction is way more important.  The rear will typically always follow, even if you get traction to break.  The front, not so much, haha.

You've shared an experiment showing your results and asking for feedback. I've shared feedback from the perspective of someone who designs experiments in a professional setting. You can disagree with the feedback as much as you want, but it doesn't change the validity of the underlying criticism of your experiment. 

The assumption about the general size of the data in this setting is incorrect. Take a look at the assumptions and mechanisms that this type of experimentation is built upon. Take a look at the differences between generative and discriminative models and their implications, what type of model is foundational for DoE? What are its assumptions? Is it a statistical model or is it machine learning? what is the purpose of replication in DoE? etc ... Don't take it from me, I'm just relaying information - there's plenty of reputable resources out there other than some internet forum.

That's great that you've enjoyed the experiment and the exercise, it was a fun read and well written. If you included blocking in your design, you'd be able to take the analysis further without adding much more cost (time, etc). For example, lets say you block, randomize within blocks, and collect the temperature at each run - you would have the ability to make statements about high/low tire pressure with respect to temperature (Does high tire pressure perform better in higher temperature vs lower for the given track? for example). You can take this further in a 2^k design to include other aspects like hi/lo shock pressure, geo, etc. One of the key reasons to use a designed experiment is the efficient use of resource - the idea is to arrive at robust inference with minimal cost incurred.

If collecting as much data as possible is something that is key to your approach and blocking and randomization is deemed not appropriate, you might want to take a look at some of the quasi-experimental techniques like interrupted times series. 

Best,

notanengineer