Reading Working Papers in Linguistics 4 (2000) - The University of ...
Reading Working Papers in Linguistics 4 (2000) - The University of ...
Reading Working Papers in Linguistics 4 (2000) - The University of ...
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
SOCIAL NETWORKS IN A RURAL SETTING<br />
<strong>The</strong> schoolchildren are mostly from the Huntly Academy, a state nonselective<br />
secondary school. <strong>The</strong> selection was made as follows: all the<br />
children fall<strong>in</strong>g <strong>in</strong>to the two age groups were given letters by the teachers,<br />
ask<strong>in</strong>g if they would participate, and those who agreed, and who obta<strong>in</strong>ed<br />
permission from their parents, were <strong>in</strong>terviewed. <strong>The</strong> adult <strong>in</strong>formants were<br />
accessed via network connections. This method makes use <strong>of</strong> the benefits <strong>of</strong><br />
association provided by network contacts (Milroy: 1988). One possible source<br />
<strong>of</strong> skew<strong>in</strong>g <strong>in</strong> the sample is <strong>of</strong> course the fact that the adult subjects were<br />
mostly accessed via network connections, but this was <strong>of</strong>fset by the fact that<br />
the research assistant’s network contacts were also brought <strong>in</strong>to play. <strong>The</strong><br />
adult sample is therefore not as ‘random’ as the child sample, which consists<br />
mostly <strong>of</strong> record<strong>in</strong>gs made at the local school, where the author had no control<br />
over the choice <strong>of</strong> subject.<br />
7. Statistical Analysis<br />
7.1. Correlations and Plots<br />
It was decided that, because the data is cont<strong>in</strong>uous, the only effective tests<br />
would be firstly correlations and then multiple (l<strong>in</strong>ear) regression, and these<br />
have been used throughout 11 . <strong>The</strong> aim <strong>of</strong> the data analysis is to see if any <strong>of</strong><br />
the social variables are correlated with any <strong>of</strong> the l<strong>in</strong>guistic ones (or <strong>in</strong> fact<br />
with any <strong>of</strong> the other social variables). This will help answer the research<br />
question related to the significance <strong>of</strong> life modes and social networks as<br />
predictors <strong>of</strong> language change <strong>in</strong> rural areas.<br />
As a first measure, Pearson’s correlation tests and plots were used to see<br />
which variables might be related to each other. More rigorous test<strong>in</strong>g was<br />
done later at the model build<strong>in</strong>g stage, us<strong>in</strong>g multiple regression. Correlation<br />
shows if two variables tend to vary together (positive correlation), or to vary<br />
<strong>in</strong> opposite directions (negative correlation). Show<strong>in</strong>g such a correlation does<br />
not, however, mean that one variable is the cause <strong>of</strong> the variation <strong>in</strong> the other.<br />
<strong>The</strong> variation may be caused by a third variable. In order to test rigorously<br />
whether one variable is caus<strong>in</strong>g another to vary, one must use regression. This<br />
will show which variable is the predictor, and which the response. A<br />
regression analysis is like the numerical representation <strong>of</strong> a scatter plot, and<br />
the regression l<strong>in</strong>e is a straight l<strong>in</strong>e through the middle <strong>of</strong> the data po<strong>in</strong>ts, one<br />
11 This was <strong>in</strong> consultation with the Applied Statistics Advisory Service at the <strong>University</strong> <strong>of</strong> <strong>Read<strong>in</strong>g</strong>.<br />
159