Long story short...it doesn't matter how the score ends up being close...you want a chance for it to be close so you have a chance to win.
What you really want is repeatability. There is no such thing as being good in close games (sample error), and if you're trying to forecast what a team will do in the future, close games are a red flag. Moreso are teams that have inconsistent or bad production relative to the final outcome. Purdue is 3-3 and has slightly negative differentials across most stats. When I see a team with a negative yards per play differential, it's an enormous red flag. (See my 2015 Sparty posts)
Purdue is making their wins seem bigger than they are and their losses seem closer than they are due to a small sample size.
Against Minnesota: they're actually down 1 with just over a minute to play. Final Score: 31-17.
Against Michigan: 10 net yards in the second half. Mich 5.64 YPP, Purdue 3.78 YPP. Final score 28-10
Against Wisconsin: Purdue 4.02 YPP, Wisconsin 6.96 YPP. Final score 17-9
One might think that they had a good chance to win vs Wisconsin, because of those 3 Wisconsin turnovers. This is true, turnovers are big, but they are not repeatable, nor are they predictable. They may have been able to win that game, but they also got outplayed by a large margin.
Purdue to date is roughly equivalent to Nebraska to date, but the perception is that they are doing much better. Hence they're being overrated.