make visualizations out of factor arrays

[maxheld83]

It seems to be customary to either a) represent factors by a list of item z-scores or b) re-create a hypothetical, ideal-typical q-sort from the position of that factor (see Watts & Stenner 2012).

There are several types of information that we should get out of this results view:

1. the significance of items on factors
2. the "loadings" of items on factors
3. the scores (ideally as continuous, z-scored data, not reduced to ordinal)
4. the "holistic" big picture in a form similar to the original q-sorts, and a form that makes sure we always understand the individual items in the context of, and relative to all other items.

I was thinking of accomplishing / summarizing all of this in a wordle-type visualization, where

• items are represented by short, meaningful titles (maybe with full item formulation available on mouse-over for a later interactive version via shiny)
• position of short-titles on x-axis is the item scores as z-scores (same as q-sort)
• y-axis is meaningless (same as q-sort)
• font size of short-names is loading
• font color (degrees from black to greyish to white) is significance

That might give a pretty good overview.

I'll into how this might be done with existing wordle etc. packages on R.

[maxheld83]

Susan Ramlo had a similar idea in 2011, though the implementation seems to use a subset of items only, and/or an actual word count of words within items, as opposed to item shorthandles.

[maxheld83]

On further thought, there are more problems/challenges/revisions:

1. item short handle font size should maybe not vary, because
   • surface of words grows quadratically whereas scores do not - > overstates magnitudes of differences
   • short handles for items will probably be of different length, thereby distorting any visualization relying on font size
   • so maybe just rely on color and/or transparency
2. x-axis positioning should be meaningful (agree/disagree obviously) (which is not the case in traditional wordles)
3. y-axis will remain meaningless (see this criticism of meaningless spatial positioning in worldles

[maxheld83]

this is the next thing I'll be working on. I feel like I really need a better way to visually inspect my results.

[aiorazabala]

Did you check the plot() for an object of class "Qmethodres" ? See pp 8 in http://journal.r-project.org/archive/accepted/zabala.pdf

[maxheld83]

yes I did! I am working with that all the time already! Thanks so much for including that! :bouquet:

I was thinking more of a visualization that basically reproduces the way data is entered, so an ideal-typical qsort for each factor, maybe in different colors, maybe with added info layers for consensus/distinguishing or something (see above). I obviously haven't thought this out.

I am working with my (diss) q study that has 77 items, so I really need a way to – as some q methodologists would stress – interpret the factors holistically.

Ps.: Have also downloaded that paper of yours, still have to read that.

[aiorazabala]

Ah, I see sorry. How about the _plot.table()_ function? Create a table where the statements are 'in the correct cell', and where there are also empty cells (to get the 'pyramid' effect), and then plot it (i.e., perhaps a function specific to Q data which is based on plot.table() )

[maxheld83]

here's a very early draft, qdc is still missing. The transparency of the tiles corresponds to the standard deviation of items for flagged q sorts.

Helps my kind of factor interpretation a lot, because I can read full items, and I can see how closely actual sorts fell around the mean.

What do you think @aiorazabala ?

[VerenaHeld]

@maxheld83 Transparency for standard deviation seems to be a good (new?) idea (and probably better than font size). It helps to keep in mind that the item positions of the "flagged/individual" q-sorts could be very different and hence more/less relevant for the interpretation...
I'm curious to know how the qdc-visualization works... maybe with little boxes (left, right or middle) at the bottom of each item (colored like the other factors)?

[maxheld83]

thanks @VerenaKasztantowicz ! Yeah; I'll have to be careful not to produce an illegibly overloaded graph ...

[maxheld83]

ok, here we go, now with (some amount of) qdc.R statements. Lines show the relative position of items on the significantly different other factor (3 factors in total). So, if a yellow line is slightly to the right of center of a card, it means that that item scored significantly more positive on the "yellow" factor. I've also fixed the layout a little bit.

There's also different fonts, but that's a different story (specific to my study).

I think this information overload, so I'll probably just make it digital, and add a little colored corner if some item is significantly more to the left or the right.

What do you think @aiorazabala and @VerenaKasztantowicz

[aiorazabala]

Beautiful!!!

[maxheld83]

@aiorazabala I'll add these as functions soon; they are essentially ggplot2 wrappers, but there's some data preparation and layouting necessary that people might appreciate to have ready-made.

[maxheld83]

ok this is pretty much the final version for now. Colors automatically cycle through a color space.

Now comes documentation, testing, input validation.

[maxheld83]

we're almost there.

• [x] improving / proofreading docs
• [x] making the legend prettier, for different kinds of nstats -> https://github.com/aiorazabala/qmethod/issues/198
• [x] check whether part duplication of qdc() code can be avoided -> #199
• [x] this function should also get a more streamlined name, like q.scoreplot() -> #201