Accidental Dwelling Fire Trends in Scotland 2009-10 to 2013-14
by Stewart Ross, Performance Data Services Manager, Scottish Fire and Rescue Service
November 2014
Added notes
- Linear trend lines are fitted for convenience and simplicity. It does not mean that the relationships shown are truly linear (which is why slide 4 provides an alternative LOESS plot version of the 32-council chart).
- The correlation co-efficient for the linear trend lines is often very low, indicating that in many cases there is really no correlation between quarterly time periods and ADFs. Slide 5 ranks the correlation co-efficients of the trend lines so you can see which of the trend lines shown may have some non-random correlation – mainly the ones I highlighted in the slides, except for Inverclyde which is just random (unfortunately there is no built-in way I can superimpose the plot coefficients on the trellis charts shown)
- Slide 6 ranks the population estimates for each local authority area, with those most of interest from the point of view of declining ADFs highlighted
- The total population of the four local authority areas showing the most decline in ADFs is about 1.3 million. The other 28 local authority areas with a total population of about 4 million show a mixed picture of in some cases increases in ADFs, some declines, but for many no overall trend at all
The SIMD ranking for each ADF was derived as follows:
- the geocode for the incident defines which of the 6505 small-area datazones the point will be counted in (and, separately, which of the 353 local authority wards in Scotland the point lies within).
- Each datazone is a bounded polygon with roughly similar population totals of about 800-1000 people. The SIMD data ranks each of these on a weighted range of indicators from 1 (most-deprived) to 6505 (least-deprived).
- The SIMD ranking for the datazone in which the incident occurred was looked up in a table populated from the 2012 SIMD base data
- The allocation of the quintile in which the ranking was counted was derived simply by banding the 6505 SIMD rankings into five equal groupings and allocating the grouping to the incident data