7. Despite all necessary prior arrangements, some respondents were
either absent on the date of interview or unable to participate for various
reason (leaving questionnaires partially filled). For this reason, in addition
to the original sample selected, a replacement sample of up-to 20% was
availed to the enumerators. The replacement sample was also randomly
generated from each cooperative’s sample frame less the selected sample.
In a few cases a second replacement sample was necessary. This was
occasioned by deaths or out migration and also due to problems with
registers of cooperative membership.
8. In the second round, 82 farmers could not be traced again and
thus dropped out of the panel sample. Annex 1 provides an analysis of potential attrition bias that appears to be mostly related to
farmers with inferior coffee production conditions, low savings, and
reduced membership of local organizations. The remaining sample
thus can be considered as an upper boundary of certification
outcomes since more marginal farmers are subject to outmigration
from the area