1 Introduction

1.2 Motivating Examples

Example 1.1.

Childhood leukaemia in Humberside, from Cuzick and Edwards (1990).

Figure 1.1: First Link, Second Link, Caption: Childhood leukaemia in Humberside with locations of cases and controls appearing as red dots. Left plot: residential locations of all known cases of childhood leukaemia in Humberside, England, over the period 1974-82. Right plot: residential locations of a random sample from the birth register over the same area and time-period. As you might expect, the cases and controls tend to be located around urban areas, but we’d like to be able to identify areas where disease rates differ.
Example 1.2.

Lung and larynx cancers in Chorley-South Ribble (Diggle et al., 1990).

Figure 1.2: Link, Caption: Lung (red dots) and larynx cancers (blue plusses) in Chorley-South Ribble. The plot also shows the location of an industrial incinerator as a black asterisk located around the center of the bottom part of the plot. The question here is whether lung and larynx cancer incidence might be related to proximity to the incinerator - this is difficult to tell by eye.

The map shows

  • •

    all known cases of lung cancer in Chorley-Ribble, England (dots);

  • •

    all known cases of larynx cancer in the same area (small crosses);

  • •

    the location of a now-disused industrial incinerator (large cross)

In Section 6 we look at a similar example in a study of asthma in children in the proximity of a coking works.

Example 1.3.

Colorectal cancer in Birmingham (Kelsall and Wakefield, 2000).

Figure 1.3: Link, Caption: Colorectal cancer in Birmingham. The map shows estimates of relative risk in each ward (legend at top of plot), adjusted for the age-sex mix of the population in each ward. Such choropleth maps are common in spatial epidemiology.
  • •

    the raw data are counts of the numbers of cases of colorectal cancer in regions Ai corresponding to 36 electoral wards in Birmingham

Substantive questions

  • •

    Do cases show a surprising tendency to cluster together?

  • •

    Does the risk of disease vary spatially?

  • •

    Is disease risk elevated near a particular location?

Testing or estimation?

  • •

    All of the above substantive questions can be expressed as hypotheses to be tested.

  • •

    But rejection of the null is only the first stage.

  • •

    We will usually want to estimate spatial effects.

  • •

    And we should ideally do so after taking account of non-spatial risk factors.