Graphics Reference
In-Depth Information
Figure . . Estimated distributions of the “gregariousness parameters” α i for women and men. he
lines drawn over the bars are posterior simulation draws that indicate the inferential uncertainty in the
histograms. Our primary goal here is to display inferences for the distribution of gregariousness within
each sex, not to compare averages between sexes (which could be done, for example, using parallel
boxplots). We compare groups using a regression model as in Fig. .
Figure . . Coe cients (and
standard error intervals) for the regression of estimated log
gregariousness parameters α i on personal characteristics. Because the regression is done on
a logarithmic scale, the coe cients (with the exception of the constant term) can be interpreted as
proportional differences; thus, with all else held constant, women have social network sizes that are %
smaller than men, people that are over years of age have social network sizes that are % lower
than others, and so forth
and
in the histogram. In the case of one scalar, we always draw the posterior intervals
that account for the uncertainty in estimation; in the case of a vector shown as a his-
togram, wesimilarly want todisplaythe uncertainty in the histogram estimate. Todo
this, we sample (say, twenty) vectors from the matrix of simulations and overlay the
twenty histogram estimates as lines on the average histograms. his gives us a rough
idea of how good an estimate the histogram is. As a rule of thumb, we don't plot the
“theta-hats” (point estimates), we plot the “thetas” (posterior draws representing the
random variables) themselves.
Search WWH ::




Custom Search