Yeah I read that but without knowing their methodology it's not useful data. For one thing, most drivers don't check the pressure or tread depth of their tires. At least some bring them in after they're completely shot. We can't say how many but it's not unreasonable to suspect that a lot of people go to a tire store because of tire damage rather than wear unless their tires are completely bald and they get a scare, someone tells them they're on unsafe tires, or they get a ticket.
Remember your methodology courses...you need at least two data points to infer a relationship. The time when customers bring their tires back in is one data point (if that's actually what the 49K average is representing) but we'd need to know what the average tread depth on those tires was, as well, to know whether they had any useful/safe tread on them when they were returned.
That is, just because customers are bringing their tires in on the 50K mark because that's what the label says they're "good" for, doesn't tell us how long they've driving on tires without usable tread. But that's just part of the problem. We don't actually know people are bringing their tires in at the 50K mark from that "average."
To understand that, we'd have to remember your stats courses: there are three different ways to measure "averages": mean, median, and mode.
Take 11 data points from customers replacing tires at 35K, 35K, 35K, 35K, 35k, 35k, 50K, 65K, 65K, 65K, and 80K. When you see "average," it's usually the mean. That's the one everyone does in their head when they calculate an average (add them all up, divide by the total). The mean in this example is 49K, but notice that only 1 customer returned the tires at 50K. That's because mean calculations are not resistant to outliers (the 80K). Remove the outlier and the average becomes 45.5K.
The median is resistant to outliers and is the average we'd want to use if we suspected the data was prone to outliers (like tire wear). The median value is 35K.
Mode can also be resistant to outliers as long as there isn't a cluster of data. Mode is the value that occurs most frequently and, in this example, is also 35K.
So now you have two calculations that customers bring their tires in, on average, at 35K miles (but we still don't know *why* they bring them in or *what* condition they're in when they bring them in) and one calculated average of 49K.
Then you look back at the data set and realize that only one customer brought tires in at 50K, though.
So what does this tell us? Well the main thing it tells us is that the expectations of the researcher as one's model is developed is going to impact the data coming out the other end. That is, if I suspect that tire wear is going to have wide variance because I assume that lots of factors impact tire wear, I'd choose the median value over mean if I wanted to ensure those outliers didn't adversely impact my conclusions. If I wanted to hide those outliers, or if I didn't expect them, I would use the mean but it wouldn't be a very good prediction of what the next person could expect.
If I was a researcher and I expected that some customers were prone to driving on their tires down to the nubs, I'd want to pull the data back to where most customers' experiences are with the mode or the middle of a normal population distribution with the median.
If I wanted to show my tires lasted a really long time and I had those same expectations (that some customers were prone to drive to the nubs), I'd use the mean in order to capture those outliers.
That's why in methods class we say things like, "garbage in, garbage out" and the public says things like, "lies, damned lies, and statistics." The numbers exist, they aren't made up, but the modeling of it shows a very different picture depending on what you assume about the population being studied and how you choose to calculate those data.