The proximal hair segment was chosen as it best correlates with the one month time frame of the diet data. Models see more in the candidate set included all combinations of the variables (e.g. Modelfish; Modelshellfish; Modelfish+shellfish; Modelfish*shellfish). [THg] was log transformed to improve normality. We examined the relationship between [THg] and δ15N and δ13C values using two separate simple linear regressions to test whether diet, as determined by δ15N and δ13C, affects mean [THg] (across segments; Proc REG; n = 73). Seventy three women had hair [THg], δ15N, and δ13C values determined. We log-transformed the data to

meet the assumption of homoscedasticity and examined for influential outliers. As we did not account for the negative sign associated with δ13C, a negative β-value indicates that [THg] decreased as δ13C is enriched (i.e. becomes less negative). Additionally, we ranked δ13C from 1 − 73 (from values of -18.52 to -12.19) and ran a regression on the ranks, AZD2281 reducing the influence of an outlying individual (δ13C = -12.19). Lastly, we used general linear models (GLM) to evaluate the relationship

between the frequency of consumption of fish and shellfish as reported by the individual and δ13C and δ15N stable isotopes values (n = 61), using 2 separate a priori candidate model sets, each with 3 models. 5-Fluoracil clinical trial Sixty one women had δ13C and δ15N measured and completed diet recalls. Models in the two candidate sets included all additive combinations of the variables (e.g. δ15N fish; δ15N fish+shellfish; δ13C fish; δ13C fish+shellfish). We added a constant (20) and square root-transformed δ13C to improve normality. Values are reported as

means ± SE unless otherwise indicated. Analyses were conducted using SAS (SAS Institute, Cary, NC). We considered results significant at α < 0.05. All statistical analyses were conducted with and without an individual with exceptionally high [THg] to ensure that this individual was not overly influential in our assessment. We used Akaike’s information criterion adjusted for small sample size (AICc) to select the best approximating models as it allowed us to evaluate a number of competing nested models without violating the rules of multiple comparisons and error rates (Burnham and Anderson, 2002). We used Tukey’s multiple comparison test to compare means. We measured [THg] in the proximal hair segments of 97 women. [THg] averaged 3.26 ± 0.97 μg g−1, ranging from 0.12 to 90.0 μg g−1 (750-fold range). When the individual with [THg] of 90.0 μg g−1 was excluded as an outlier, [THg] averaged 2.35 ± 0.38 μg g−1 and ranged from 0.12 to 24.20 μg g−1 (202-fold range).