Of course the lower the sample size the more cautious you need to be about extrapolating from that, but it is just wrong to say there can never be any significance in low sample sizes. If your wife falls off the hotel balcony and dies on your honeymoon, that is very sad. If your second wife also falls off the hotel balcony and dies on your second honeymoon, that suggests a pattern.
It depends entirely on context. If you are conducting an opinion poll on voting intentions, a sample of fifteen respondents is a very small number on which to base any assessment. If you are looking at what is happening to one team during a 38 game season, 15 games is not a bad sample size at all.
But if you want to look at the time added on during the earlier part of the season, fill your boots. I don't have the time. What I will say is what I've already said - those who have suspected that less time is added on when we are chasing a win are right. You can reject any suggestion that this is deliberate, or otherwise argue about the reasons for that all you like, but the numbers don't lie.