Results from Class 2005

The mapping exercise started well with lots of good data, but in the end the map seemed pretty hard to interpret. Obvious features such as fuel smells along roads with perhaps some natural odours on the river, but somehow the data seemed poorly represented by the mapping technique adopted. It was argued if only we had GIS all would have been OK! Not everyone was convinced, though that this was a rapid solution to our problems.

Suggestions for next year

1. Less intensity classes - combine 1 and 2
2. Give a handout on Chi-square
3. Update the street map of Norwich
4. Question members of the public at various points - possibly not prompting them with the actual odour classes - see how they think the site smells..
5. Offer a web based system so data could be typed in prior to the practical
6. Do it in the computer lab with GIS facilities

These seem good suggestions and the first four/five seem quite easy, although people may find the calculations more difficult on a computer. Excel is quite weak at non-parametric statistics and students may not know other software. Similarly not also students are familiar with GIS software.

General review of the data structure

The odour intensities and the classification of odour type can only have integer values. Although the intensity values are ordered, the category of smell is non parametric data. You cannot get the average smell of Norwich, by averaging the categories! The mode (most common value, ie will work for class of odour) and median (central value of an ordered series, will work for intensities) are useful parameters. We see in Fig. 1 that the smell of fuel is the most typical type of odour in Norwich.
Intensities We need to choose these kinds of distributions which are not sensitive to statistical distributions and the non-paramateric nature of the data. This inolves such considerations as: Normal distribution of data Non parametric and ordered integer data Use of mode and median Contingency tables and Chi-square analysis Results from Class 2002 We can see in the figure to the right that the intensity of odour has a distribution over the range 1-5. Unfortunately I forgot to get the 0 values from the class on Friday. We can see the modal intensity is 2. This is interestuing to compare with earlier years on a much less windy day where the mode was intensity 3 although someone that year disobeyed the rules and registered an intensity of 6! I tried a binomial distribution which was not especially satisfying without the zero values, so I have plotted these in the figure to the right without details on the calculation. The data fits only poorly to a binomial distribution (agreement can be rejected at the 99% level - I combined classes 4 and 5, so had only 3 degrees of freedom). We can see that our observations over-represent weak smells (intensity 1)... perhaps a further indication of the windiness of the day(?)
Results from Class 2002 Analysis of the difference bewtween the earthy-floral and the fuelly-burnt showed a significant difference at the 95% level (in fact p=0.022). The results are available as an Excel file.
Earlier Results on different odours One of the questions that concerned a previous workshop was the statistical difference between the distribution of floral-fruity plus putrid-fishy-earthy and fuel intensities. It was interesting here to note that the food and floral odours were quite different in their distribution, with observations that the food observations became more frequent at higher intensities, while less so for floral odours. Indeed the modal intensity was 1 for floral odours. A Chi-squared analysis, based on the table below, yeilds a value of 3.78 which with two degrees of freedom suggests that the two data sets are different at better than the 10% level. Odour I=1 I=2 I>2 Floral 8 5 4 Food 4 6 10
A parallel interest is the statistical difference between the distribution of smoky-acrid-burnt and fuel smell intensities. Certainly the indications are from a visual inspection is that there is a difference. It is particularly noticeable that the fuel odours, although most common tend to be rated weaker than other odours. Note how the modal intensity for fuel is 2, while in other classes such as burnt, solvent or food it is 3. Is this simply psychological and a function of rating more common odours as weaker ones?

Re-designing the experiment

There was a feeling we could have an earthy-algal category to replace earthy-putrid-fishy. Some suggested care was needed in integrating the time and intensity elements into assigning the intensity of the smell. Variability and windiness were a problem. This would be helped by making an observation on the site and an integrated one on the walk between sites.

General discussion

Norwich 50 years ago would have had a more noticeable odour of coal smoke. Then 200 years ago dung, rotting food and sewage would have been more characteristic. There are numerous biases in the observations, but these probably reflect real difficulties in treating reported odour complaints.