Something’s wrong with those numbers. Medians of integer-valued quantities are always integers or half-integers.
EDIT: I’ve taken a look at the report, and it doesn’t say anything about how they calculate medians, so I don’t know how they’re fudging their numbers to get these out.
EDIT 2: I should also say “good job for looking at the research and getting numbers”, even if I’d like these researchers to be more transparent as to what they’re actually reporting.
It’s almost certainly true, perhaps doing a weighted average of the medians of subgroups. However, any method that does that is not producing a median. A good way of doing that adjustment might give “cooked” numbers for the various options, but the point where 50% are below and 50% are above would still almost certainly be an integer. And if it is actually balanced (highly unlikely with so many data points), so that any number greater than X and less than X+1 divides the population in two, then the convention is to report X + 1⁄2. There is no information about the median that anything past the decimal point can actually convey.
Something’s wrong with those numbers. Medians of integer-valued quantities are always integers or half-integers.
EDIT: I’ve taken a look at the report, and it doesn’t say anything about how they calculate medians, so I don’t know how they’re fudging their numbers to get these out.
EDIT 2: I should also say “good job for looking at the research and getting numbers”, even if I’d like these researchers to be more transparent as to what they’re actually reporting.
An uninformed guess: those medians are presumably based on survey data, so they might’ve been adjusted using the survey’s sampling weights.
It’s almost certainly true, perhaps doing a weighted average of the medians of subgroups. However, any method that does that is not producing a median. A good way of doing that adjustment might give “cooked” numbers for the various options, but the point where 50% are below and 50% are above would still almost certainly be an integer. And if it is actually balanced (highly unlikely with so many data points), so that any number greater than X and less than X+1 divides the population in two, then the convention is to report X + 1⁄2. There is no information about the median that anything past the decimal point can actually convey.