The apparent fraud with Research 2000 polls made me think about the way that polls are conducted. Polling is so much of an art form and the profession is full of secrets when it comes to the methods a pollster uses that it is almost impossible for any outsider to verify the results. So in every poll, the pollster's name and reputation is the only reason we trust their work or not, and we shouldn't really have had to wait for a scandal like this to know that trusting people with a totally unverifiable product based on their name alone is a BAD idea.
So let's look at the profession, and look for ways in which we can improve the transparency of the polling process, and add further checks to the process to ensure more trustworthy polls.
Split the job into two parts: Data Gathering and Analysis
The main financial incentive for a fraud on R2K's part - if any such fraud occurred - is to save on the costs on gathering data. Gathering a sample size of 1200 people means they would have to make about 4000 calls, since most people hang up on pollsters, and that takes a lot of people and a lot of calls which in turn raises your costs. So let's make sure that there are two companies involved...
It would bring so much more trustworthiness if you contract two different companies. The company that Analyzes the data will have to define the questions asked, and when the data is available they will be in charge of weighting it according to their formulas and models, but a whole other company is in charge of making the calls and gathering the raw data.
It would be incredibly hard for a fraudulent pollster to generate a bad dataset that could fool the analyst part of the poll. An analyst part of the poll will have very few incentives to cut corners, and in case anyone finds a way to do it anyway, the raw data is available from the data gatherer to check.
Is there an angle I've missed here, what's the argument for NOT splitting the job up, besides a possible increase in cost?