An analysis of roadside vegetation condition ratings using Markov Chains
by
Suzanne Brown
Murdoch University
Coauthors: Dr Ross Taplin (Murdoch University)

We consider an interesting applied statistical problem based on data relating to the condition of roadside vegetation, a factor important to the prevention of erosion and maintenance of the natural environment. Attempts to quantify the roadside vegetation condition along Western Australian roads lead to data with strong spatial dependence but with several assessments from different raters acting independently.

We briefly explain alternative ways of approaching the problem of obtaining an overall profile of the roadside condition (ie, the proportions of the road assessed in each rating category), and explore in detail two approaches. The first calculates a profile for each rater and statistically averages these independent estimates. The second approach calculates a median "consensus" rating for the road, which is then analysed using Markov models.

The decision of whether to analyse independent profiles or to analyse the median sequence has implications with regard to the estimates and standard errors of the proportions obtained. These, together with some of the practical constraints imposed by the users of the data and the possible cost savings of using a systematic sample to rate only a fraction of the roadside, are discussed.

Date received: March 18, 2002