PMC:4101712 / 29688-31209 JSONTXT

{"target":"https://pubannotation.org/docs/sourcedb/PMC/sourceid/4101712","sourcedb":"PMC","sourceid":"4101712","source_url":"https://www.ncbi.nlm.nih.gov/pmc/4101712","text":"The above results, often called a main effects analysis, can be improved by adding interaction terms into the regression. An interaction term between factors i and j adds an additional regression term of form ξ i,j f i (s) f j (s) into equation 3. Table 3 reports a logistic regression model fit with most non-mosquito factors allowed to pairwise interact. The notation Temperature*Elevation, for example, refers to temperature and elevation interaction. There are 82=28 possible interacting pairs. However, only 16 interaction pairs were considered due to practical constraints. For example, interaction between elevation and temperature is plausible since heartworm prevalence at higher elevations may differ according to temperature. However, median household income and temperature could not be sensibly allowed to interact since heartworm prevalence for dog owners with high salaries does not depend on the temperature where he/she lives, and vice-versa. The mosquito factors were not allowed to interact with other factors because they are simply presence/absence variables. All insignificant factors and interactions were eliminated at the 5% level with a standard backward elimination regression procedure [25]. Clarifying, we first fitted a model with all individual factors and 16 interactions. The term with the largest p-value in the regression was eliminated if its p-value exceeded 0.05 and the model was refitted. This procedure was repeated until all insignificant factors were eliminated at the 5% level.","tracks":[]}