Much of the dialogue surrounding effective pedagogy for college-level general education astronomy courses has been focused on how best to engender  “science literacy”, yet students cannot be scientifically literate without also being quantitatively literate. Basic numerical skills such as graph reading, proportional reasoning, and estimation are tools that we as scientists are trained to use as reasoning tools, and they benefit us in our daily lives as voters, consumers, and citizens. I will discuss results from the Quantitative Reasoning for College Science (QuaRCS) assessment, an online multiple-choice assessment administered at the start and end of a semester of general education college science instruction and designed to measure numerical skill and affect. The QuaRCS has been administered to more than 10,000 students at dozens of institutions over the past ten years. I will discuss themes that have emerged from this large dataset, including evidence supporting three broad conclusions : (1) numerical self-efficacy, a perception of math’s relevance to daily life, and math anxiety level are strongly predictive of student score on the assessment,  (2) achievement gaps for various demographic groups narrow substantially by compensating for these affective variables, and (3) the words and cultural contexts used in assessment questions affect their accessibility to students, impacting performance. These conclusions are important in their ability to inform effective curricular practices in general education science instruction, especially in regard to helping students improve numerical and science literacy levels. They are also an important part of the puzzle in the work that we all must do to reduce barriers to entry to STEM disciplines faced disproportionately by students from historically underrepresented groups.