Graphics Reference
In-Depth Information
Figure
.
.
Profile plot of a single five-day auction. he circles represent bids, with circle size
proportional to the bidder's eBay rating
Standard visualization tools are geared towards the display of either time-series data
alone or cross-sectional data alone - almost never both.
he combination of time-series and cross-sectional data into one display is rare
andrequirescareful,application-specificmodificationsofstandardmethods.Shmueli
&Jank (
)proposetheuseof profile plots for displaying the temporal sequence of
bidstogetherwithadditional auction attributes (suchasabidder'srating) inthesame
graph. his is illustrated in Fig.
.
, which describes the sequence of bids in a five-
dayeBayauction. hecirclesizeisproportional tothe bidder'seBayrating. However,
profile plots are more suitable for visualizing single auctions, and do not scale well.
Another type of plot that is suitable for visualizing functional data is the rug plot
(Hydeetal.,
).Arugplotdisplayscurves(i.e.,functionalobjects)overcalen-
dar time in order to explore the effects of event concurrency. Figure
.
shows a rug
plot displaying the price curves of
eBay auctions of a Palm M
PDA that took
place over a three-month period.Each colored line represents an individual auction,
and it is estimated via monotone smoothing (Ramsay and Silverman,
). Mono-
tone smoothing is computationally more expensive than penalized smoothing, but
it ensures that the resulting line increases monotonically. he black line represents
the average daily closing price. We can see that daily prices vary quite significantly,
and so does the daily variation in price (gray bands). Furthermore, we can see that
there are time periods with many similar, almost parallel price curves for the same
auction durations (e.g., seven-day auctions - green curves - at around
/
and also
around
/
). Moreover, the closing prices ater
/
appear to be relatively low, and
so does the associated price variability. Most auctions closing at that time are seven-
day auctions with similar shapes. It would be interesting to see if one could establish
a more formal relationship between similar price patterns (i.e., parallel price curves)
and their effect on the price and its uncertainty.
he rug plot in this example combines functional data with attribute data via the
time axis (calendar time on the x-axis takes into account the start and the end of
