Re: Praise for RepGrid2 and a question about its cluster analysis
Brian Gaines (gaines@cpsc.ucalgary.ca)
Thu, 24 Jun 1999 21:31:37 -0700
Tony Downing asks:-
>
>2) Question about RepGrid2 cluster analysis
>===========================================
>
>The Repgrid2 cluster analysis procedure is called "Focus". When you do a
>Focus analysis, there is a prelimiinary screen with various boxes to check.
>Two of them, which are alternatives, are labeled "Edges" and "Interior". I
>got the impression that these offered a choice between alternative
>clustering algorithms, with "Edges" probably selecting the Single Link
>Algorithm and "Interior" probably selecting some other algorithm such as
>the Mean Link Algorothm. Trying out checking one or the other of these
>boxes, I find that they certainly do make a difference, but I have been
>utterly dumbfounded to find that choosing one or the other option, with an
>already-elicited grid, changes some of the the actual ratings of particular
>elements on particular cnstructs.
>
>Clearly I have misinterpreted the function of these check-boxes! Can
>anybody enlighten me?
>
>(In case the program is misbehaving, I should perhaps say that we are using
>21-point scales. The participant will not see these numbers until we show
>them the grids after Time 2, but the choice of 21 levels ensures that when
>an element-label is dragged to its position along the construct-line, it
>stays where it is put - it doesn't snap to a coarsely-quantised pasition.)
>
>But what do "edges" anmd "interior " mean?
The FOCUS algorithm sorts the grid to bring similar elements together and
similar constructs together. In bringing constructs together it may reverse
one or more so that the left hand pole becomes the right hand pole and vice
versa. In doing do it also reverses the ratings so that, in your data, 1
becomes 21, 2 20, etc.
The default 'Edges' algorithm sorts the grid according to Mildred Shaw's
original FOCUS algorithm (described in detail in Shaw, 'On Becoming a
Personal Scientist', Academic Press, 1980) which incrementally links the
highest matched item to the edges of existing clusters. The variant
'Interior' uses matches to all existing items in an existing cluster,
rather than just the two outer edges. It was developed as a check on the
stability of the FOCUS clusters and is not recommended for routine data
analysis.
b.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%