Closed maxheld83 closed 8 years ago
Here we go.
Pretty close to the standard plot that @aiorazabala build, but some extra bells and whistles.
I find the interpretation to be harder than q.scoreplot()
, probably because the text is harder to read / relate to the data, but there's now way around that (overplotting).
as became evident in fixing #293, the difference between
zsc_n
andzsc
can actually be quite large.Consider this example:
Here, looking at
zsc_n
, the itempoor-central-planner
, looking fromresentment
, appears to be quite assymmetrically distinguishing, with a difference of-2
and4
, respectively. However, when looking at the originalzsc
, we see that in fact, the qdc diff is quite symmetrical.I was stumped by this in double-checking the qdc lines (which are based on
zsc
), though this is of course a bigger issue that relates to allzsc_n
.zsc_n
naturally implies a loss of information, though troublingly, a particular kind: it will (typically) enforce a kind of spread that's not really there in the factor scores.Of course, if
zsc_n
and its plots are strictly interpreted as ordinal data, there's no problem: it is correct.We must be careful however not to interpret the horizontal dim as distance, which it is not.
Obviously, the other big problem here is that a continuous kind of plot would lead to horrendous overplotting.