Proc freq

PROC MEANS and PROC FREQ in SAS version 9.3 (SAS Institute Inc) were used to generate means and percentages to describe the samples of schools, principals, and teachers. Policy knowledge among principals and teachers was compared with the content of the written school district sun safety policy, and principals’ and teachers’ knowledge of each component was classified as accurate or inaccurate. The effects of role (principal vs teacher), demographics characteristics of the principal or teacher, and characteristics of the school on awareness and knowledge of school district sun safety policy were assessed by using multilevel analysis, with individuals (principal or teacher) nested within schools and schools nested within districts. PROC GLIMMIX was used to fit mixed models for logistic regression on awareness of policy and PROC MIXED was used to fit mixed models for linear regression on number of policy elements correctly known. We set α criterion levels at .05 (2-tailed).

A chi-square (PROC FREQ, SAS 9.3) test was used to analyze data from the first (four-chambered) olfactometer experiments. To identify if preferences existed among the odours emanating from imbibed and unimbibed seeds of the three weed species, residence time for each weed seed species and a control (four-choice assay) were compared with mixed-model ANOVA (PROC MIXED, SAS 9.3) [33 ], using residence time as the response variable to be predicted by weed species. Analyses were carried out separately for both imbibed and unimbibed seeds for each carabid species, with weed species treated as a fixed effect.
Data about imbibed seeds did not meet the assumptions of ANOVA; therefore, we fitted generalized estimating equations (GEEs) to these data using PROC GLIMMIX (SAS 9.3) with a negative binomial error distribution function (PROC GLIMMIX) [33 ]. Numerous possible error distribution functions were tested, including the Poisson distribution, but the negative binomial error function provided the best fit (based on the ratio of Chi-square/df ratio being close to 1). We tested models similar to those for unimbibed seeds.
Differences in mean residence times among weed species were compared using Tukey’s post-hoc test. Data from the second bioassay comparing responses to unimbibed and imbibed B. napus seeds were analyzed using a chi-square test (PROC FREQ) to establish if preferences existed.

The total number of Ae. albopictus numbers from each state were analyzed separately for this study, because the experimental set up was not consistent among states. Adult Ae. albopictus counts (total number of females and males) from BGS1 and BGS2P traps were compared using negative binomial regression with a log link (PROC GENMOD, SAS version 9.4 for Windows) for all experiments except for data from New Jersey (Experiment 1), and Florida (Experiment 4). Overdispersion in the New Jersey model was not adequately accounted for by the negative binomial distribution, therefore a quasi-Poisson model, scaled using the Pearson chi-squared statistic divided by the degrees of freedom, was used instead. For all models, the total number of adults was regressed against time and trap type. Time was modeled as a continuous linear covariate for all comparisons; however, in Experiment 1, we used piecewise linear regression with a knot at day 22 to account for curvature at the end of the trapping period. All pairwise comparisons were examined using Holm’s correction to control for multiplicity. The Florida data were analyzed using chi-squared or exact chi-squared tests (PROC FREQ, SAS version 9.4 for Windows).

Statistical analyses were done using JMP version 12 and SAS version 9.4 for Windows (SAS Institute; Cary NC). Factorial analysis of variance (ANOVA) was performed on continuous variables from the BAL fluid and the log fold changes from the qPCR analytes to make comparisons between the treatment groups. For some variables, data were log-transformed to reduce heterogeneous variance and meet the assumptions of an ANOVA. Dunnett’s test was used for individual comparisons to sham. Histopathological findings using the graded scale were analyzed using nonparametric Kruskal Wallis tests followed by Wilcoxon Rank Sum tests for pair-wise comparisons. Gross tumor counts and histopathology counts from sections were analyzed similarly. Post hoc comparisons using the nonparametric Dunn method for joint ranking were conducted to compare the treated groups to the sham group, and to compare low dose to high dose within each component. Lung tumor incidence was analyzed using a Chi-square test in SAS ‘Proc Freq,’ while tumor multiplicity was analyzed using Poisson regression in SAS ‘Proc Genmod’. In cases where overdispersion existed, a negative binomial regression was performed using data from those animals surviving to the 30-week time point. For all analyses, a p-value of <0.05 was set as the criteria for significance.

We determined whether similar sets of genes have an elevated dN/dS and pN/pS ratio by running a correlation analysis on the log transformed dN/dS and pN/pS ratios from each gene or concatenation (PROC CORR, SAS 9.3, SAS Institute, INC. 2011). Only genes or concatenations that had three or more SNPs were included in the correlation analysis (n = 11).
Additionally, to test whether genes with elevated dN/dS ratios have been under positive selection versus relaxed purifying selection, McDonald-Kreitman (M-K) tests [12 (link)] were run using Fisher’s exact test for each set of concatenated or individual genes (PROC FREQ, SAS 9.3, SAS Institute, INC. 2011). Pairwise divergence data between C. americanum and each of the three outgroup species (T. caeruleum, H. annuus, and N. tabacum) were obtained using a Perl script to extract all pairwise SNPs from the four species gene alignments used in the plastid divergence analysis above (Additional file 4). We applied a Bonferroni correction factor of 51 to account for multiple comparisons (17 genes/concatenations × 3 outgroup species).