The main form of exposure was defined by two variables: the
proportion of farms using pesticides and the proportion of farms
reporting cases of pesticide poisoning. Although they both indicate
exposure to pesticides, they reflect a different intensity of
exposure. Thus, their effects are not independent and they cannot
be included in the same model. Considering the epidemiological
controversy about the relationship between pesticide exposure
(without poisoning) and rates of suicide, as well as pesticide
poisoning as an important indicator of intense exposure, two
models were developed, one for each main exposure variable.
All of the above mentioned variables were identified in the
official databases that record agricultural work and production, as
were the socio-economic and demographic aspects. According to
the literature, these aspects might be associated with suicide and
were available for the micro-region’s aggregation level.
To examine qualitative variables (as religion, skin colour and
marital status) and to include them as continuous variable in
multiple linear regression, it was necessary a dummy variable,
examining if a characteristic was present or not. Only one variable
was selected to represent each indicator. Priority was given to the
measure with strongest correlation with suicide rates out of all the
possible measures for that variable. Therefore, the dummy variable
used for religion was ‘% of Catholics’, and the indicator of skin
colour was ‘% of white-skinned people’ (Szklo and Nieto, 2012). For
marital status, the indicator was ‘the rate of divorces and other
legal separations/100,000 inhabitants’ following the literature
(Faria et al., 2006).