Gah! I’m trying to sort hotels in Las Vegas, for which Trip Advisor is critical – how else would I know, for example, that the reason the Monte Carlo is cheap at the moment is there’s a big construction project going on in their car park, with overnight pile driving?
However, the traveller ratings are still tricky to interpret – take the Imperial Palace, for example. I’ve found a fantastic deal, but is it any good? Well… Trip Advisor’s reviews vary greatly. It’s clear that some people have had an awful time, but does an average 3/5 rating mean it’s broadly OK, or so variable it’s risky? 3±1 is more reassuring than 3±2.
If more people understood the maths, it’d make perfect sense to quote standard deviations for this sort of thing. To get around that, I wonder if we should start drawing a distribution curve something like Tufte’s sparklines above those five star rating things. That would convey how grouped the ratings are, right?
3 thoughts on “Standard deviations”
My experience with trip advisor is you’ll always get a couple of miserable sods but any more than a couple of negative comments then avoid. The people who were happy with a hotel that lots of others are calling crap probably have low standards… “Hmm lovely mild instant coffee and plenty of milk in the scramble egg powder – just how we like it dear.”
Of course: some people will always have a bad time, and others will have different tastes. But this is why having an idea of the shape of the distribution is so important. If there are only a dozen reviews then it’s reasonable to read them all — but for Vegas hotels there are usually 500 or more. I want to know what proportion give the place low marks, as well as the average.
I’d like to see National Rail enquiries quote some standard deviations (compiled from their live train monitoring site) on their timetable site. A useful rail journey planner would estimate your chances of making your connection, given the performance statistics, and calculate your backup plan. The return leg of my regular Aberystwyth commute (now over) would often involve missing a +31min connection in Birmingham, then making a -1min connection in Derby, arriving home at the originally proposed time. There are lies, damn lies and statistics. Just occasionally, we could do with more statistics.