Seth Masket, a political scientist, is pretty typical of the profession in the tools he uses to make predictions about electoral outcomes and other political realities.
For example, he notes that Presidential approval ratings have little to do with legislative accomplishments and a lot to do with the state of the economy.
He is likewise skeptical that late breaking information should change his electoral predictions. He bases that prediction on Presidential approval ratings on Labor Day and nine month GDP growth rates at around the same time.
There are other models of electoral outcomes by political scientists, but most of them share of the common feature of giving little or no weight to the quality of the campaigning that goes on. They look at who controls the Presidency, how many seats controlled by each party usually vote for the other party, the number of open seats, whether it is a midterm or non-midterm election, voter registration changes since the last election, unemployment rates and similar factors.
Most political science models implicitly assume that the quality of campaign season rhetoric, the amount of money spent by 527s, the laws passed in the last couple of years, and so on, have a relatively modest effect on electoral outcomes.
If these models are basically right, then campaign finance regulation may not be that important because campaign ads purchased with that money may not have much of an impact on the outcome. It may be that money poured into races is caused by strong candidate support, rather than the other way around (as there is no doubt that winning candidates tend to have raised more money, and hence there is a correlation, even if there is not a strong line of causation, between campaign spending and electoral results).
Of course, most political science models also have high margins of error, and campaign finance may be the "God of the Gaps" that determines where in the range of possibilities outcomes end up.
But, there is lots of competition for that role. Late disclosure of key information can dramatically swing elections (e.g. the disclosure of Scott McInnis's plagerism just before the primary election). Candidate recruiting clearly matters as well - conservative districts won by Democrats, for example, are overwhelmingly won by conservative Blue Dog Democrats. Third party spoilers can also skew outcomes, something that is critical to Hickenlooper's lead this year and which was an important factor that helped elect Bill Clinton. Particularly back in the days when elections were conducted exclusively on election day, weather on the day of the election impacted turnout, often with decisive results. Newspaper endorsements and free media coverage play a part. Perhaps the stances that candidates take on issues and "foot in mouth moments" do matter somwhat.
To the extent that some or all of these factors matter, the share of the margin of error in political science models of electoral outcomes attributable to the actual conduct of the campaign shrinks.
This doesn't mean, of course, that spending doesn't matter at all. But, it does mean that if each candidate spends what they can raise in typical years, that the results more or less cancel out.
Put another way, theoretically, it is possible to distinguish between people who would vote that same way with or without campaign advertising and those who changed their vote because of campaign advertising. If the outcomes don't depend much upon campaign ads, then the number of people who changed their vote because of campaign advertising is presumably quite small. Thus, the amount of campaign spending per vote swung may be an order of magnitude or two more than the amount of campaign spending per vote cast. In dollar terms, this means that it may cost something on the order of a hundred dollars or more to change a single marginal vote.
If true, perhaps the immense amount of effort that goes into trying to make elections more fair via campaign finance regulation is mispent. For example, maybe offering pre-paid stamps on ballots and mandatory voter registration and all mail in ballots would have more impact at leveling the playing field at a comparable cost. Maybe allowing candidates to make brief statements in the "Blue Book" would counterbalance immense volumes of much more expensive campaigning. Maybe making election day a holiday would have a bigger impact.
In a constitutional order where we place a premium on the Freedom of Speech, especially in the political context, these alternatives are attractive, not just from a cost perspective, but because they don't regulate speech in the same way that campaign finance regulation does.
Of course, on the other hand, if campaign finance does indeed have a big impact and the prevailing political science models have more to do with convenience than accuracy, then perhaps greater regulation is in order.