The Network Scale-up Method (NSUM) uses social networks and answers to "How many X's do you know?" questions to estimate hard-to-reach population sizes. This paper focuses on two biases associated with the NSUM. First, different populations are known to have different average social network sizes, introducing degree ratio bias. This is especially true for marginalized populations like sex workers and drug users, where members tend to have smaller social networks than the average person. Second, large subpopulations are weighted more heavily than small subpopulations in current NSUM estimators, leading to poor size estimates of small subpopulations. We show how the degree ratio affects size estimates, provide a method to estimate degree ratios without collecting additional data, and demonstrate that rescaling size estimates improves the estimates for smaller subpopulations. We demonstrate that our adjustment procedures improve the accuracy of NSUM size estimates using simulations and data from two data sources.