Yes, Fairgen could reliably augment all 18 niche segments by at least 2x using predictive synthetic data.

In order to conclude that, we've used a public available survey of 18,000 rows investigating political preferences of UK residents. Fairgen was trained on a small fraction of it (1,000 rows), and the sub-segments of interest were analyzed.

The median was very low, below 2%, and the quartiles were close together, showing that Fairgen could remove edge cases. It performed better than a real sample twice as big as the trained one, showing that Fairgen could effectively predict and augment each niche by at least 2x.

When looking at 1 under-represented niche “Green Party electorate”, Fairgen could reliably augment all 161 questions while maintaining low median (right above 2%) and quartiles close together.

No problem! We also compared the performance of the augmented dataset on one specific question, of one specific sub-segment.

Here we see that Fairgen could successfully predict, being much closed to the ground truth than the real sample.

Political research teams can expect to gain faster granular insights from their quantitative efforts. It is clear that synthetic data can be used to reduce data collection while still meeting weighting variables, especially when the segments of interest are increasingly niched. The impact is faster and more granular insights, that can drive quick deicision-making.