explanatory variable in a context of interaction with hard programs.^{11 }equation takes the form:

The regression

ε β β β β + + + + + = k k x x x y . . . 2 2 1 1 0

(7)

# Where:

y = is dependent variable, in this case vehicle trip rate at worksite x_{1 }, x_{2 },..., x_{k }= explanatory variables (soft and hard program policies, firm characteristics,

other controls) ε = stochastic or error term

Equation (7) can include higher order term to acknowledge nonlinear relationships, and interaction terms between the response variables.

# Given that the objective is to build a model that can predict the effect on vehicle trip rate

reduction of consecutive

one or more soft program

years

are

needed.

In

the

initiatives, worksite trip reduction data between analysis, the Washington Sate Department of

Transportation Trip Reduction Program dataset was obtained for the period 1995 The data reports information on worksite characteristics, such as firm size and type, employee mode share, and information of TDM programs.

to 2005. industry

The data were analyzed and factor analysis was employed to reduce the number of explanatory variables to improve model prediction power.^{12 }During the model building phase, several variations of Equation (7) were considered. At the end, a predictive model that allows for interaction between qualitative variables was chosen as the one with the

higher predictive power.^{13 }predictive model to be used of this report^{14}.

A table of diversion rates was developed based within the sketch planning tool described in the next

on this section

11 The model herein proposed to build upon previous work conducted by CUTR in estimating worksite trip reduction tables [30].

12 Factor analysis is a statistical technique that reduces several variables that are correlated into a smaller set of new, uncorrelated and meaningful variables.

13 In a regression model, qualitative variables take the form of dummy variables. These are explanatory variables that take the value of 1 if present or take the value 0 if absent. For example, dummy variables can be used to estimate main effects due to the presence or the absence of a given program promotion initiative, a given subsidy, and the offering or not of a guaranteed ride home program. Furthermore, very often these initiatives are linked to each other in an interactive fashion. An interaction model has to be built to analyze a main effect model.

14

The diversion rates accompany the model CD.

38