• No results found

L IMITATIONS OF THE RESEARCH

CHAPTER 8 EVALUATION AND CONCLUSIONS

8.3 L IMITATIONS OF THE RESEARCH

This section looks into the limitations of the methods for modelling the spatio-temporal movement of tourists. Limitations regarding survey design and implementation are also discussed.

8.3.1 Limitations of the modelling methods

The MC theory is used extensively in statistical modelling to link a sequence of events together under the assumption of first-order dependence (that is, the future evolution of events is only dependent on its present state). In this thesis, MC methodology was utilised to analyse the outcomes and trends of events associated with tourist spatio-temporal movement patterns.

It was a novel method for modelling the spatio-temporal movement of tourists at the macro level. From the output of the model, it was found that as the number of attractions visited in a sequence of movement gets larger the MC model breaks down and therefore, does not fit the data well. In this case study, the performance of the MC model is still good because there are only seven destinations for tourists on the daily trip. However, if there is a longer sequence of movement higher-order Markov Chains should be considered.

One interesting result from the case study is that the Markov property is not reliable in explaining tourist movement behaviours under certain circumstances. For example, tourists who visited the KCC will probably not visit it again on a single day trip. The MC models cannot accommodate this situation. A suitable model should assign zero or small probabilities to tourist returning to attractions that have already been visited.

One important issue for utilisation of Log-linear models is sample size. When the sample size is small, the accuracy of the residual and component chi-square statistics will be affected adversely (Kennedy 1992). In particular, when the frequencies of movement patterns are zero, the log of zero will be −∞ . Consequently, the likelihood ratio Chi-square G becomes 2 meaningless. Goodman (1984) suggests adding 0.5 to the observed frequencies of movement patterns of tourists. The package SPSS adopts this method. Agresti (1990) recommends that an extremely small constant (such as 108) can be added to the cell frequencies. Both methods can overcome the difficulties with the chi-square. But if zero is recorded too often as the frequency of movement patterns, the expected frequencies of movement patterns will converge to zero during iterative fitting. Therefore, a big sample size is necessary (Kennedy 1992).

The EM algorithm implements a statistical model to identify tourism market segments for each significant movement pattern of tourists. If appropriate, an optimal result will be achieved from the algorithm. However, the lack of explanation of tourism market segments is one of the limitations of the EM algorithm. Sometimes, expert experience or knowledge is needed to identify the characteristics of tourism market segments. Witten and Frank (2000) suggest that a supervised model, namely, the classification method, could be used to analyse the results outputted from the EM algorithm.

In this thesis, variables such as tourist spatial and social abilities, their knowledge of the region they are visiting, their levels of familiarity with the environment, their individual

motivation, their spatial and temporal constraints and the configuration of the physical environment were used to build the wayfinding decision-making models. However, there could be other variables, such as physical limitations of tourists, which this thesis did not consider. Therefore, inclusion of more variables in the cognitive model can optimise the model, which is another direction for the future research.

8.3.2 Limitations of the survey design and implementation

Two surveys were conducted from 6-8 March 2004 and from 17-20 January 2005 on Phillip Island. The survey included nine attractions as tourist destinations on this island. Four attractions: the Penguin Parade, the KCC, Churchill Island, and the Phillip Island Information Centre were chosen as sampling locations. As mentioned in section 7.3.1, most questionnaires were collected from the Penguin Parade. Therefore, nearly 5% of tourists who did not visit these four attractions could have been missed by the surveys. (Particularly, domestic tourists who preferred to stay elsewhere or to surf or swim at beaches during their visit).

Another limitation of the case study is the sampling time. Both surveys were conducted in the high season. Therefore, low seasonal tourist behaviour was not tracked. During the low season, especially winter, 30-50% fewer tourists visit Phillip Island and the number of attractions that the tourists visit during their daily trip could be smaller (Phillip Island Nature Park 2001-02). Comparison of movement behaviour between the high and low seasons could be the subject of further research in the future.

Finally the sample size is questionable. Different types of data analysis need different sample sizes. For example, 464 self-administered questionnaires for tracking movement at the macro level, and 132 movement records tracked by GPS along with questionnaires at the micro level were collected from Phillip Island. The sample sizes at both levels are suitable for most analyses conducted in this thesis. However, for data mining methods, especially for classification, which is used to identify the differences between tourist groups, a larger sample size is required (Witten and Frank 2000). Therefore, in this thesis, the classification method was not adopted.