Destinations

  Once the probability of embarkation and length of trip are determined in the course of the simulation, the boat next needs to select a destination from a list of all possible destinations. The probability that a destination would be selected was determined from the boater interview data based on relationships between kinds of boats and kinds of destinations preferred by the interview subjects.

In the boater interviews, subjects provided free-form lists of places where they liked to take their boats in what they described to be typical trips. The following steps were used to map boat types to likely destinations:

1.

create common list of destinations: Each free form description was scrutinized, and a list of destinations was extracted from the sentence. This step is necessarily subjective, as it requires interpretation from the person scrutinizing the list to determine which locations the subject is really talking about in the list. This was partially mitigated by the fact that all entries were interpreted by a single person. Example: Subject: ``I take my boat out the canal to the ICW, sail north to Longboat Pass, and fish in the flats for a while. Then I pull up the anchor and head to Sister Keys to hang out on the beach for a while.''

This becomes: ``subject: ICW; Longboat Pass; flats; Sister Keys''

2.

arrange data into tables:  The data then were translated into tables, with the standardized destinations as the columns and the respondents as the rows or entries. A value of 1 was placed in the columns where the respondent identified the corresponding place as a typical destination, and 0 in columns that were not mentioned. This portion of the interview was not standardized, so it was not considered practical to read in relative rankings of destinations based on the interview data. Some people naturally volunteer more information than others; there is no realistic way to say that the racing sailor who always takes his boat out to a set of racing buoys (a single destination) is more or less enthusiastic about boating than the pontoon boater who mentions twelve destinations that s/he frequents. These produced large matrices: there were 122 boaters who identified themselves as winter boaters who provided free lists, with a total of 90 destinations that they frequent. There were 112 boaters who identified themselves as summer boaters, with a total of 72 destinations that they frequent (appendixB).

3.

statistical analysis: Two primary methods were used to infer natural groups of boats, destinations, and associations between kinds of boats and kinds of destinations. Given the form of the data in the tables described in step 2, it was impossible to establish statistically significant correlations between boat types and destination types based on the raw data. However, useful inferences still can be seen in how data were clumped together; certain groups of boats and destinations were consistently grouped together using the two techniques, and the groups made intuitive sense (destinations that were in fishing flats tended to get grouped together, and cruising destinations inside and outside the bay were consistently grouped together). Given the low scores of statistical significance, the results of these tests can only be considered as an aid to conceptual groupings of boats and destinations, and cannot be seen as conclusive or relevant for any information or analysis beyond this parameterization step.

A principal component analysis (PCA), which is a powerful variable reduction method, was used to analyze both the covariance and correlation matrices of the matrix formed in step 2. Since the correlation of two variables is calculated directly from the covariance, it is natural that the two analyses should agree in their clustering of variables; however, both methods yielded insights into how the variables cluster together, and will be discussed in more detail here.

Additionally, the Euclidean distance between observations based on variable responses, and between variables based on observations was calculated. These distance matrices were used as measures of psychological distances between variables and observations, and a multidimensional-scaling (MDS) algorithm was used to render these distances into ``mental maps'' of the boating population and boating destinations. MDS provided additional insights into similarities and conceptual groupings of destinations.

 

 

|t:=========:t| Destination 2|cSail 2|cSpeed 2|cRec. Fish 2|c||Power Cabin

Prob. Cum. Prob. Cum. Prob. Cum. Prob. Cum.

||-----|| Bay 0.38 0.38 0.15 0.15 0.09 0.09 0.10 0.10

Gulf 0.38 0.76 0.15 0.30 0.09 0.18 0.10 0.20

New Pass 0.03 0.79 0.06 0.36 0.11 0.29 0.05 0.25

ICW 0.02 0.81 0.06 0.42 0.07 0.36 0.12 0.37

Flats 0.02 0.83 0.06 0.48 0.11 0.47 0.05 0.44

Group I 0.03 0.86 0.06 0.54 0.04 0.51 0.12 0.56

Group II 0.02 0.88 0.06 0.60 0.11 0.62 0.05 0.61

Group III 0.03 0.91 0.06 0.66 0.04 0.66 0.12 0.73

Group IV 0.02 0.93 0.09 0.75 0.11 0.77 0.05 0.78

Group V 0.03 0.96 0.06 0.81 0.04 0.81 0.12 0.90

Group VI 0.03 0.99 0.09 0.90 0.11 0.92 0.05 0.95

Group VII 0.02 1.01 0.06 0.96 0.07 0.99 0.05 1.00

||-----|| Total 0.97 0.96 0.99 0.98 |b:=========:b|

 

 

|t:=========:t| Destination 2|cSail 2|cSpeed 2|cRec. Fish 2|c||Power Cabin

Prob. Cum. Prob. Cum. Prob. Cum. Prob. Cum.

||-----|| ICW and Gulf 0.21 0.21 0.11 0.11 0.07 0.07 0.40 0.40

ICW and Bay 0.06 0.27 0.05 0.16 0.15 0.22 0.05 0.45

Bay 0.41 0.68 0.15 0.31 0.17 0.39 0.10 0.55

Flats 0.00 0.68 0.03 0.34 0.09 0.48 0.03 0.58

Group I 0.02 0.70 0.10 0.44 0.09 0.57 0.05 0.63

Group II 0.07 0.77 0.07 0.51 0.06 0.63 0.05 0.68

Group III 0.04 0.81 0.10 0.61 0.09 0.72 0.03 0.72

Group IV 0.04 0.85 0.07 0.68 0.09 0.81 0.08 0.74

Group V 0.02 0.87 0.07 0.75 0.09 0.90 0.03 0.82

Group VI 0.04 0.91 0.07 0.82 0.03 0.93 0.08 0.90

Group VII 0.04 0.95 0.07 0.89 0.03 0.96 0.08 0.90

Group VIII 0.02 0.97 0.10 0.99 0.02 0.98 0.02 0.97

||-----|| Total 0.97 0.99 0.98 0.97 |b:=========:b|

Details of the analysis, interpretation, and resulting tables are presented in appendix A. Tables 4.11 and 4.12 show the probabilities derived from this analysis which were used in calibration of this aspect of the simulation. Note that some destinations have membership in more than one group.