Adequate sample size across the range of each variable is a concern when testing interactions. Table 1 provides descriptive information on peer smoking and the SNP rs16969968. We found no difference in the distribution of H.S. peer smoking by genotype of rs16969968 (χ2= 11.26, df=16, p=0.79). Additionally the smallest cell size included 15 subjects (AA genotype with one smoking peer). Thus it appears unlikely that the interaction of H.S. peer smoking and rs16969968 identified in these analyses is an artifact of small cell sizes. We also tested a model in which H.S. peer smoking was treated as an ordinal variable. These analyses found a statistically significant interaction between H.S. peer smoking and rs16969968 comparing the lowest to highest quartiles of number of smoking peers (p = 0.002). Thus the interaction finding appears robust to both the distribution of variables and scale used to model the peer smoking variable.
