what are "matched pairs" from a statistical point of view?

Answer 1

They are pairs in which the members of each pair are matched in some sort of reasonable way, in some way related to whatever it is that you're testing. For example, suppose you wanted to compare the effectiveness of 2 different diets. And suppose you decided to do that by having 2 groups of people, with each group using a different diet.

Maybe you'd put the 2 groups on different diets and measure their weight loss in a few months, to see which diet seemed to work better.

Well OK, then, in order to do that, you wouldn't want one group to be all old fat people and the other to be all young skinny people, would you? No, of course, you'd want them to be comparable -- and you might know or guess that both age and starting weight could have an effect on their weight loss.

So what you could do is have "matched pairs": Select 2 groups with the same number of people, and select them so that each pair of people represented the 2 groups (one person in each of the diet groups) and so that the members of each pair were of about the same age and weight --that is, matched in terms of their age & weight. ... so the members of one pair, for example, might both be about 30 years old and both weigh about 130 pounds, and another pair, might both be 50 years old and both weigh close to 180 pounds.

Then if one group did better than the other, you could say it apparently wasn't due to age or weight, because those factors had been controlled for, in your matched pairs design.

You could also, if you thought it important, match them in terms of sex or ethnic group or whatever ... in order to make the results comparable, & so as to give you satisfactory evidence that any different results in their weight loss appeared to be due to their diets, and not age or sex or whatever you matched them on.