Odour Threshold Investigation 2012 - Bay of Plenty Regional Council
Odour Threshold Investigation 2012 - Bay of Plenty Regional Council
Odour Threshold Investigation 2012 - Bay of Plenty Regional Council
You also want an ePaper? Increase the reach of your titles
YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.
14<br />
Histogram <strong>of</strong> MnH2S_Whakatāne<br />
12<br />
10<br />
Frequency<br />
8<br />
6<br />
4<br />
2<br />
0<br />
0<br />
MED<br />
GMEAN<br />
MEAN<br />
1<br />
2<br />
MnH2S_Whakatāne<br />
3<br />
4<br />
Figure 7<br />
Distribution <strong>of</strong> Whakatāne thresholds.<br />
We now (finally) give the results <strong>of</strong> useful analyses comparing locations which show the clear<br />
difference in average thresholds between the locations. Although our full analysis used data<br />
with all results per session per panellist, it is sufficient to simply compare the log thresholds<br />
at the two locations using a two-sample t-test. The test is justified by the approximate<br />
normality <strong>of</strong> the logged form <strong>of</strong> the data. And given that there are only two locations it is also<br />
simple to allow the variability at the locations to be different.<br />
Two-sample T for log10MnH 2 S<br />
Location N Mean StDev SE Mean<br />
Rotorua 29 0.159 0.291 0.054<br />
Whakatāne 30 -0.208 0.362 0.066<br />
Difference = mu (Rotorua) - mu (Whakatāne).<br />
Estimate for difference: 0.3671.<br />
95% CI for difference: (0.1960, 0.5381).<br />
T-Test <strong>of</strong> difference = 0 (vs not =): T-Value = 4.30 P-Value < 0.0005 DF = 55.<br />
We have shown there is very strong evidence that the geometric mean thresholds at Rotorua<br />
are higher than at Whakatane. Our point estimate <strong>of</strong> the difference in the log scale is 0.3671<br />
which corresponds to a ratio in geometric means <strong>of</strong> 10 0.367 = 2.3. We are 95% sure that the<br />
ratio is somewhere between 10 .196 and 10 .538 i.e. between 1.6 and 3.5. Note that the<br />
estimated geometric means are 10 -.208 =0.62 and 10 .159 = 1.44 as listed previously.<br />
Estimating other useful characteristics <strong>of</strong> the threshold<br />
distributions<br />
As an alternative to using transformations, some statistical s<strong>of</strong>tware <strong>of</strong>fers alternative<br />
approaches to leaving the data in the untransformed state and estimating various percentiles<br />
(such as the median), the mean, and also the proportion <strong>of</strong> the data in various data ranges.<br />
The results below used such a procedure and included fitting the threshold data with a threeparameter<br />
lognormal distribution (equivalent to the data transformation Y = log e (threshold –<br />
0.16)).<br />
Environmental Publication <strong>2012</strong>/06–A review <strong>of</strong> odour properties <strong>of</strong> H 2 S <strong>Odour</strong> <strong>Threshold</strong> <strong>Investigation</strong> <strong>2012</strong> 67