Geography Reference
In-Depth Information
agronomists) and the survey was directed to producers of i ne wines. Data were gathered
using the universe of i ne wine producers populating the three clusters, comprising 32 in
Colline Pisane, 41 in Bolgheri/ValdiCornia and 32 in Valle de Colchagua, summing up
to a total of 105 i rms. The data were collected using a structured questionnaire, which
allowed relational data to be collected through a 'roster recall' method so that each i rm
was presented with a complete list (roster) of the other i rms in the cluster, and was asked
to report about their relations with other i rms (see the Method section).
Method
This chapter adopts multiple regression quadratic assignment procedure (MRQAP),
originally developed by Mantel (1967) and Krackhardt (1987), to test whether business
and knowledge networks inl uence a similarity matrix of i rm performance, which is used
here as a proxy of the evenness in the performance of cluster i rms. This method consid-
ers the general model:
Y
= a + b
1
X
1
+ b
2
X
2
+ e
where
Y
is a square (network) matrix of dependent observations,
X
1
is a square matrix
of observations on the independent variable of interest, and
X
2
is a square matrix of
observations on a second independent variable of interest, a is a constant term, and e
is a square matrix of residuals. The MRQAP evaluates the signii cance of b
1
and b
2
in a
dif erent way from conventional econometrics. Since the problem with statistical analysis
of social network data is the lack of independence of dyadic observations (Laumann and
Pappi, 1976), the MRQAP permutes the data many times, re-estimating the b
1
and b
2
coef-
i cients under each permutation, and then comparing the observed b
1
and b
2
coei cients
to the null hypothesis reference distribution created by the set of re-estimated b
1
and b
2
.
The specii c method adopted in this chapter is based on double Dekker semi-partialling
procedure, developed by Dekker et al. (2005) and performed through UCINET 6.116.
Dependent variable
The dependent variable is represented by a similarity squared
matrix of i rm performance. The matrix is composed of
n
rows and
n
columns, corre-
sponding to the number
n
of i rms in the network. Each i rm has a row and a column,
which are labelled 1, 2, . . .
n
. Each cell in the matrix reports the existence of a similarity
existing in the performance of i rm
i
in the row and i rm
j
in the column. In this specii c
case, in the matrix, there is a 1 in the (
i
,
j
) cell if i rm
i
and i rm
j
are
both
good performer
i rms. Cells (
i
,
j
) report a 0 when either both i rms are poor performers or at least one of
the two is. The matrix that results from this is called
Good performer
. A i rm is considered
here a 'good performer' if either of two criteria are met: (1) one of the i rm's wines has
been rated as having good quality in the international wine journal
Wine Spectator
in the
period 2003-05; (2) the i rm's export share is higher than the cluster average.
The i rst indicator (1) adopted to measure the good performance of i rms (
Rating
)
is drawn from
Wine Spectator
.
9
The wine rating of this journal is based on the quality
assessment of an international panel of expert oenologists, who review more than 12,000
wines each year in blind tasting. After tasting, oenologists assign a score to each wine
brand according to a 100-point scale, ranging from 100, when the wine is of outstanding
quality, to 50 when it is of poor quality. A set of information is listed with each rated
