8 JANUARY/FEBRUARY 2018 • FOGHORN FOGHORNFOCUS: SALES & MARKETING Figure 2. Forecast model of passenger and vehicle ridership with underlying data – Guemes Island Ferry Figure 1.Annual vehicle round trips plotted over time – Guemes Island Ferry with ridership from 1980-2002, indi- cating that it was a likely driver of the ridership pool. The county popula- tion dataset was the closest correlate to island community population that was available consistently since 1980, and also included decades of official forecasts that were essential for ridership prediction. After 2002, however, ridership no longer seemed to be directly coupled to population growth. Skagit County had surmised that fare increases were a possible cause of declining ridership. Others thought that ridership had stopped increasing because the ferry had reached full capacity. Glosten theorized that these drifts were the result of rational consumers reacting to other factors that we had not yet included in our ridership model. Our primary goal was to under- stand the influence of factors that our client could control, such as fares and parking. However, we also wanted to understand the influence of factors that our client could not control, such as economic cycles. Analysis and Methods We gathered data to test a range of hypotheses. Skagit County provided fare history, which we adjusted for inflation. Skagit County also provided schedule history, from which we measured the number of runs and the hours of service offered per week. From a variety of sources, we estimated the number of parking spaces at both terminals over time, which we divided by the county population to get a better sense of the relative availability of parking. We also gathered economic data, such as unemployment rates and housing indexes. In order to develop a ridership model that could account for the influence of more than one factor, we made a multivariable linear regres- sion. This statistical technique fits a line through one series of output data based on multiple series of input data. Input series are only kept in the model if they satisfy three conditions. They must be correlated with the output data; they must improve the correlation between