Environmental Engineering Reference
In-Depth Information
specify the number of groups that your data should divide into (here, the four
participatory wealth classes). The algorithm then divided the data into four
groups in the way that that minimised the variation within the groups and
maximised the variation between them [see web links at the end of the chap-
ter for more information on the method].
●
The correlation between the cluster that a household was placed into, based
on quantitative wealth measures, and the participatory wealth group it
belonged to could then be assessed. The authors found a correlation coeffi-
cient of 0.26, which with 121 datapoints was significant at P
0.01. This
means that there was still a fair amount of scatter, but that there was clearly a
relationship between the two measures of wealth. Notice of course that the
qualitative and quantitative wealth measures are not entirely independent, as
the variables in the quantitative measure were identified qualitatively. But the
statistical analysis gave the authors confidence that both methods were telling
them the same thing, and so it was possible to use either as a valid representa-
tion of wealth. Note too that the fact that a particular household may be in
different wealth classes under different methods is not a problem so long as
overall the methods give similar results, and the differences between them are
not systematically biased in one direction or another.
●
Three continuous
monetary variables
were also calculated for each house-
hold: production (the market value equivalent of crops and wild food pro-
duced, plus gifts and net profits from sales), consumption (the market value
equivalent of all the goods consumed) and sales (market value income from
sales of crops and wild foods). These variables were standardised to take into
account differences in household size and composition, being expressed in
units of US$ per adult male equivalent. This was important in order to ensure
consistency within the sample. Note that wealth was measured in monetary
units; this allowed more complex statistical analysis to be carried out, at the
potential expense of a deeper qualitative understanding.
●
Next, the variables, season and wealth rank, were used as factors in a
gener-
alised linear model
of production and consumption. GLMs have the same
underlying philosophy as any other regression model. They are available as
standard in many statistical software packages. However, it is not straightfor-
ward to produce a correctly specified model and then interpret the output.
Crawley (2005, 2007) provides excellent guidance on how to carry out GLMs
using the free software R.
The approach to data analysis was highly quantitative, despite the participatory
way in which the data were collected. Several different approaches were taken to
presenting the data. This included using means to give a straightforward expres-
sion of the differences between groups. For example, the poorest wealth group
produced goods worth only US$0.1 per adult male per day, while the richest group
produced US$0.72 per day—still well below the UN definition of extreme poverty
as living on
$1 per day. They also used flow diagrams to demonstrate the way in
