Table 2.
Linear regression modeling of the relationship between insulin and covariates
| Model | Parameters |
F value 2 degrees of freedom |
P value for ethnicity × BMI |
Parameter estimate for ethnicity (black) × BMI |
Parameter estimate for ethnicity (Mexican- American) × BMI |
|---|---|---|---|---|---|
| Model 1: Insulin = ethnicity + BMI + ethnicity × BMI | |||||
| 1 | Ethnicity × BMI interaction term | ||||
| Women | 14.23 | <0.0001 | −0.010 | 0.005 | |
| Men | 4.54 | 0.01 | −0.002 | 0.008 | |
| All of the below modeled for women only | |||||
| 2 | Model 1 + diabetes (binary) | 5.09 | 0.006 | −0.021 | 0.021 |
| 3 | Model 1 + education (binary, <high school vs. >high school) | 11.94 | <0.0001 | −0.011 | 0.007 |
| 4 | Model 1 + % calories from carbohydrates (highest quartile vs. rest) |
8.54 | 0.0002 | −0.009 | 0.003 |
| 5 | Model 1 + age (in 20-year age groups) | 11.72 | <0.0001 | −0.018 | −0.001 |
| 6 | Model 1 + income (family income <$20,000 vs. >$20,000 | 12.63 | <0.0001 | −0.010 | 0.000 |
| 7 | Model 1 + physical activity (no physical activity vs. any) | 16.32 | <0.0001 | −0.019 | 0.006 |
| 8 | Model 1 + family history of diabetes | 11.71 | <0.0001 | −0.007 | 0.021 |
Modeling in SAS. Models 2–8 contain covariate and all first- and second-order interaction terms. Interaction term contains 2 degrees of freedom. BMI was modeled continuously.