As discussed in another essay (The "Illusion" that Team Events are More Reliable than Pair Events), the outcome of Swiss teams better reflects ability than the outcome of a pairs event. However, this could be an illusion -- this fact alone gives us no reason to think that Swiss teams better reflect ability that day.
In the absence of hard data, and assuming anyone cares about the reliability of team versus pair events, this essay considers other mathematical factors that influence reliability in teams and pairs events.
At match points, there are some flat boards that don't count for much. But at IMPs, there are a lot of boards that don't matter much.
Mathematically, weighting some boards more heavily than others is actually good if the more heavily-weighted boards are more reflective of ability. One could argue that slam hands are more reflective of skill than other hands, because there is a lot of skill in bidding hands.
But I think it would be tough to argue overall that the important boards at IMPs are more reflective of skill. For example, some hands depend only on a finesse. There is no skill in playing those hands. I believe that situation is more likely to occur in a slam than a part score. Similarly, there is no defensive skill in just following suit, but that is more likely to occur in slam hands.
This fits my experience. I have played in IMP pairs and matchpoint pairs, and the matchpoint pairs seems a lot more reliability because more boards are relevant.
In board-a-match, the comparison is between just two tables. So the board-a-match corresponds to the two-table game in terms of reliability.
Put another way, in a board-a-match, the outcome for you at your table might be a score of 60%, if you were scored against the whole field. The outcome for your teammates at the other table might be 50%. That gives you an advantage of 10%. On the next board your score might be 100% against the field and your teammates might be 80%. That gives you an advantage of 80% on that board. In board-a-match scoring, these are both converted to a plus score -- you score 1 on both boards.
This is discarding useful information. Discarding useful information always decreases reliability. If this information was used -- your scores for the boards were 10% and 80%, the outcome of the event would be much more reliable. (This would be a more controlled version of my "coupled pairs", discussed in the other essay.
This again fits my experience, though I have played in only one board-a-match.
In fact, they do not. The early rounds tend to be much less important than later rounds, and the difference between the first round and the last round is huge. This is because teams with equal scores are paired. This means that if you are doing well, you compete against better teams than if you are doing poorly. This means you are essentially punished for doing well in the early rounds and rewarded for doing poorly in the early rounds.
Parenthetically, there is no easy solution to this problem. Pairings of teams could be random, but this substantially increases the chance factor of ability of your opponents. I do not know which procedure is more reliable. One could post hoc give each team a correction factor based on the ability of the opponents (as reflected by the opponent's final score.) This is similar to a computer rating of football teams. This might be very interesting; it might also violate the spirit of the Swiss Team competition.
A multi-day pairs event has the same problem. Suppose it is a three-day event. Mathematically, reliability is highest when the three days of a three day event are weighted equally. However, carry-overs are typically truncated, making the third day more important than the second and the second more important than the first.
Again, this can be justified if the last day is a more accurate reflection of ability. The last day typically has a more homogenous field, which probably increases reliability. My intuition is that this increase in reliability of the last day does not come close to compensating for the truncation. I suspect that the truncation is done to increase the interest of the last day, which is pretty close to saying it is done to decrease reliability.
This is a chance factor. All events have it. Certainly, your outcome at Swiss Teams, knockouts, or board-a-match is influenced by the quality of your opponents. In this case, a random selection of 13 pairs from the field will tend to be reasonably close to average most of the time.
The problem comes in the second session. Ideally, in the second session you would not compete against any of the pairs you competed with in the afternoon. However, there is no way for all of the pairs to compete against all new pairs. A practical solution is a random seating for the second session. If there were three or four sections, you would expect on average to be competing against just one pair for the second time. (A slightly better procedure would be to control the second seating assignment to keep the number of "rematches" the same for everyone.)
The worst possible procedure, from the standpoint of reliability, is to have you compete at night against the same pairs you competed against in the afternoon.
This worst possible procedure seems to be common at ACBL tournaments.