Table 5.1 shows the results of the two MNL models and the heterogeneous heuristic model (HHM). Comparing the two types of models, the best model in terms of CAIC is the MNL model with logged variables. The log-likelihood (LL) of HHM is the highest, but the complexity of the model is much higher. The optimal HHM turns out to have two thresholds for tR and three thresholds for iL The pedestrians seem to have mentally represented tR into three states [<70 min, 70—240 min, ≥240 min) and represented tA into four states [<14:30, 14:30—16:00, 16:00—20:00, >20:00). These segments are quite reasonable and conform with people's habit of using typical clock hours as decision references. The positive weights mean that as time goes by, the value of going home increases, but not in a linear fashion. The negative β° suggests that decision strategies with strict judgment standards are preferred by the pedestrians in general. As a result, it is less likely that pedestrians decide to go home early during the trip so that they may have more opportunities to enjoy shopping.

Three states for relative time and four states for absolute time mean that the number of preference structures is 13 = 3*4+1. Each preference structure implies two heuristics, one starting from processing tR and the other starting from processing tA. The probabilities of these 26 heuristics are estimated; the results are shown in Figure 5.1. In the Figure, the larger the index for a preference structure, the higher the overall threshold. The general trend is that the probability increases as the standard becomes stricter, due to the negative β°, implying that simpler rules are preferred. The probabilities drop at Ф11 and Ф12 because they imply risky heuristics with high probabilities of rejection. However, although Ф13 implies one of the most risky strategies – unconditional rejection, its probability is still high because its processing effort is 0.

The distributions of the two factor processing sequences differ very little before Ф6, which means when the judgment standard is low (i.e. under which the pedestrians are more prone to go home), the factor processing sequence does not matter too much. While after this point, the importance of search sequence increases and tA becomes the first factor to process most of the time. Excluding the 'no action'

Table 5.1: Estimation results of the go-home models (ENR).

MNL normal variables

MNL logged variables

Parameter

Estimate

Parameter

Estimate

βκ

-0.005e

βκ

-0.869e

βΑ

-0.004e

βΑ

-4.402e

βΗ

-6.006e

βΗ

-35.424e

Nc

808

Nc

808

Np

3

Νρ

3

LL

-415

LL

-410

CAIC

854

CAIC

843

HHM

70 min

<5?

14:30

240 min

<?2

16:00

1.000e

cA

20:00

0.766e

W2

0.822e

βε

-2.690e

nj

0.710e

F

4.526e

2.566e

β°

-1.026e

Nc

808

Np

8

LL

-396

CAIC

853

aThresholds are not counted as free parameters as only their corresponding weights potentially have an effect.

bParameters in ( ) were set for the estimation. One value parameter is set to 1 because only the relative relationships between the values matter.

cParameters that are significant/effective. Only these parameters are counted for calculating CAIC.

Figure 5.1: Distribution of preference structures (go-home).

heuristics implied by Φ1 and Φ13, the probability of tA to be searched first is 62% in total, while tR has a probability of 18%.

Found a mistake? Please highlight the word and press Shift + Enter