. Author manuscript; available in PMC: 2022 Dec 1.

Published in final edited form as: Proc Int AAAI Conf Weblogs Soc Media. 2022 May 31;16(1):228–240.

Table 6:

Average predictive accuracies (Pearson r) across four tasks when using adaptive binning.

Baseline = .640	Minimum Count Threshold
Baseline = .640	1	10	50	100	1000

Age	.583	.605⁺	.624⁺	.636⁺	.634⁺
Gender	.639	.639	.639	.640	.640
Income	.612⁻	.666^*	.674^*	.663^*	.642⁺
Education	.648^*	.648^*	.648^*	.647^*	.642
Age + Gender
Naive	.580	.598⁺	.622⁺	.633⁺	.633⁺
Raking	.580	.603⁺	.623⁺	.635⁺	.634⁺
Inc. + Edu.
Naive	.612⁻	.659^*	.673^*	.662^*	.643
Raking	.611⁻	.662^*	.674^*	.664^*	.643
All
Naive	.634	.633	.620⁻	.634	.645⁺
Raking	.579⁻	.610⁺	.634⁺	.649^*	.647^*

⁺ and ⁻ indicate a significant increase or decrease, respectively, as compared to the same correction variable / method pair in Table 5

increase over baseline. “All” includes age, gender, income, and education. This method shows mitigating selection bias can improve predictive accuracy when adjusting for error in demographic scores by using adaptive binning.