in the Son La province. First about 20 percent of the communes per
district were picked, and then one village per commune and 15
households per village were randomly selected.7 However, in two
districts of Son La, research permission was denied due to the sensitive
border location ofthe villages and other security issues. Some
households could not be located due to migration. Migration had
been prompted mainly by the construction of a huge hydroelectric
dam. For instance, one survey village had migrated in its entirety
and we were not able to track down those households. Our sample
is therefore somewhat smaller (−25 percent households) than the
VHLSS for Son La.
In Thailand a representative sample of households was drawn in
the Chiang Dao district, Chiang Mai province in northern Thailand,
using a three-stage random sampling procedure. First half of all
communes in the district of Chiang Dao, then half of all villages
per commune and then ten households per village were selected.
Both country surveys were divided into two phases (with a 3–4-
month time interval). Two reasons played a role in the adoption
of this procedure: 1. an extensive interview in which the respondent
had to answer all the questions in one interview might have
been perceived as too burdensome by the respondent; 2. some of
the data from the first interview were used as input for the second
round. Due to the short time lag between the two survey rounds,
we also experienced a low level of attrition (less than 5 percent)
caused by migration, death, and refusal. The two survey rounds covered
all information concerning social capital and social networks
of the households, as well as information on income and household
assets. As this paper focuses on formal and semiformal credit,
we used only a sub-sample of our data and restricted our analysis
to households having a formal or semiformal credit arrangement.8
For the analysis we selected all formal and semi-formal loans that
had matured within the last year before the interview and those
currently outstanding. We excluded loans taken within the last six
months before the interview because as usually a grazing period
applies in this time frame. Outstanding loans older than one year
were labelled as not performing and added to the credit history.
After excluding households with missing values, the Vietnam sample
consists of 198 loans in 171 households and the Thailand sample
of 467 loans in 225 households.
3.2. Personal network data collection
We used the name and position generator to measure personal
networks and to create measures of individual social capital. These
are well established survey tools in sociology but are rarely applied
in development economics. The name generator was developed by
McCallister and Fischer (1978), the position generator by Lin and
Dumin (1986). Both generators have been used for data collection
to measure social capital (Lin, 1999a,b). The name generator asks
questions about certain domains of the personal network, such as:
‘Whom can you ask to help you fix your car?’ Then the name of this
person is recorded. Later, more questions can be asked about that
person, for instance to ascertain the person’s sex, age, occupation,
and so forth, or to establish the relationship of this person to the
respondent. This part ofthe survey is called ‘name interpreter’.9 The
name generator has often been criticised for being biased towards
strong ties, because the first names that people recall are usually
those of persons who have been known to them for a long time, or