12.01.2015 Views

RESEARCH METHOD COHEN ok

RESEARCH METHOD COHEN ok

RESEARCH METHOD COHEN ok

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

584 MULTIDIMENSIONAL MEASUREMENT<br />

and Goldstein 1991). This has been done using<br />

multilevel regression and hierarchical linear<br />

modelling. Multilevel models enable researchers<br />

to ask questions hitherto unanswered, e.g. about<br />

variability between and within schools, teachers<br />

and curricula (Plewis 1997: 34–5), in short<br />

about the processes of teaching and learning. 4<br />

Useful overviews of multilevel modelling can<br />

be found in Goldstein (1987), Fitz-Gibbon (1997)<br />

and Keeves and Sellin (1997).<br />

Multilevel analysis avoids statistical treatments<br />

associated with experimental methods (e.g.<br />

analysis of variance and covariance); rather<br />

it uses regression analysis and, in particular,<br />

multilevel regression. Regression analysis, argues<br />

Plewis (1997: 28), assumes homoscedasticity (where<br />

the residuals demonstrate equal scatter), that the<br />

residuals are independent of each other, and<br />

finally, that the residuals are normally distributed.<br />

The whole field of multilevel modelling has<br />

proliferated rapidly since the early 1990s and<br />

is the basis of much research that is being<br />

undertaken on the ‘value added’ component of<br />

education and the comparison of schools in public<br />

‘league tables’ of results (Fitz-Gibbon 1991; 1997).<br />

However, Fitz-Gibbon (1997: 42–4) provides<br />

important evidence to question the value of some<br />

forms of multilevel modelling. She demonstrates<br />

that residual gain analysis provides answers to<br />

questions about the value-added dimension of<br />

education which differ insubstantially from those<br />

answers that are given by multilevel modelling<br />

(the lowest correlation coefficient being 0.93<br />

and 71.4 per cent of the correlations computed<br />

correlating between 0.98 and 1). The important<br />

point here is that residual gain analysis is<br />

a much more straightforward technique than<br />

multilevel modelling. Her work strikes at the<br />

heart of the need to use complex multilevel<br />

modelling to assess the ‘value-added’ component<br />

of education. In her work (Fitz-Gibbon 1997: 5)<br />

the value-added score – the difference between a<br />

statistically-predicted performance and the actual<br />

performance – can be computed using residual<br />

gain analysis rather than multilevel modelling.<br />

Nonetheless, multilevel modelling now attracts<br />

worldwide interest.<br />

Whereas ordinary regression models do not<br />

make allowances, for example, for different<br />

schools (Paterson and Goldstein 1991), multilevel<br />

regression can include school differences, and, indeed<br />

other variables, for example: socio-economic<br />

status (Willms 1992), single and co-educational<br />

schools (Daly 1996; Daly and Shuttleworth<br />

1997), location (Garner and Raudenbush 1991),<br />

size of school (Paterson 1991) and teaching<br />

styles (Zuzovsky and Aitkin 1991). Indeed Plewis<br />

(1991) indicates how multilevel modelling can be<br />

used in longitudinal studies, linking educational<br />

progress with curriculum coverage.<br />

Cluster analysis<br />

Whereas factor analysis and elementary linkage<br />

analysis enable the researcher to group together<br />

factors and variables, cluster analysis enables<br />

the researcher to group together similar and<br />

homogeneous subsamples of people. Thisisbest<br />

approached through software packages such as<br />

SPSS, and we illustrate this here. SPSS creates<br />

adendrogramofresults,groupingandregrouping<br />

groups until all the variables are embraced.<br />

For example, here is a simple cluster based on<br />

20 cases (people). Imagine that their scores have<br />

been collected on an item concerning the variable<br />

‘the attention given to teaching and learning in<br />

the school’. One can see that, at the most general<br />

level there are two clusters (cluster one = persons<br />

19, 20, 2, 13, 15, 9, 11, 18, 14, 16, 1, 10, 12,<br />

5, 17; cluster two = persons 7, 8, 4, 3, 6). If one<br />

were to wish to have smaller clusters then three<br />

groupings could be found: cluster one: persons 19,<br />

20, 2, 13, 15, 9, 11, 18; cluster two: persons 14, 16,<br />

1, 10, 12, 5, 17; cluster three: persons 7, 8, 4, 3, 6<br />

(Box 25.22).<br />

Using this analysis enables the researcher to<br />

identify important groupings of people in a<br />

post-hoc analysis, i.e. not setting up the groupings<br />

and subgroupings at the stage of sample design,<br />

but after the data have been gathered. In the

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!