> In article <9408231553.AA14302@paradox.psych.columbia.edu>,
> dhk@paradox.Psych.Columbia.Edu wrote:
>
> Let's make this discussion traceable, electronically. I'll volunteer to
> serve as an intermediary for all traffic, if that's acceptable to both
> parties. Dr. Velleman, please e-mail your data to StatSoft's e-mail
> account. If you'd rather it be kept completely public, post the data to
> sci.stat.edu (or a listserver that StatSoft can get directly). StatSoft
> (Pawel or Tom, presumably), will you respond accordingly (either cc: me on
> e-notes to Dr. Velleman or post to appropriate groups)?
While dcm's proposal may sound reasonable, in fact it cannot work. The
issue is not "data". It is first and foremost the nature and tone of
Statsoft's negative advertising. Second, is the fact that this advertising
contains many errors of commission and omission. If there were but one or
two, I would have offered to correct them, but in a document such as this I
would virtually be rewriting the entire piece. Let me provide just one
example from the third paragraph on Data Desk. (My comments will be in
parens.)
"For example, there is no way to label values of specific variables (false,
any text variable can be used to label values, or several can be used
together), no way to move blocks in the spreadsheet (false, all standard
Mac editing works, including moving blocks of data) there is, in fact, no
data spreadsheet there, data can be edited only via "scrollable lists"
(that's not what they are called) of values. (true, Data Desk does not use
a standard spreadsheet for editing, but the description here makes this
sound like there is no way to edit tables of data, when in fact Data Desk
simply uses an alternative choice of method for editing data that offers
all of the capabilities (and more!) of a spreadsheet) Also, the
user-interfact for all numerical methods neglects the *exploratory* aspects
of data analyses. (A nonsensical statement, and one judged false by most
reviewers) For example, you cannot *explore* (note the snide itallics) your
data by using a continuous variable for grouping or coding ... unless you
previously create a new variable that contains the codes. (False. Data Desk
offers derived variable expressions that include If/Then/Else decisions.
Moreover, cutpoints can be set by graphically controlled sliders. Results
update automatically and immediately both in the coded variables and in all
analyses in which they are used.) You cannot interactively specify
intervals (false), select a subset of values for codes (false), or request
that a variable be split into a reqquested number of ranges (false) e.g. to
cross tabulate a variables (Data Desk does cross tabs differently and, I
think, more conveniently and more appropriately than that.) All of these
options appear to be rudimentary requirements for interactive *exploratory*
(snide itallics again) data analysis, and they are supported even (snide)
in programs that unlike Data Desk do not claim to specialize in EDA. (I'll
appeal to the readers, is this a useful objective comment?) There is no
data transformation language included in Data Desk (false) even simple
formulas cannot be executed repeatedly in loops (There is no need, Data
Desk's transformations are vector-oriented.) There are also other
deficiencies of data management in Data Desk. e.g., the program cannot
import Excel worksheets (true, but not needed. You can copy and paste any
part of a spreadsheet from any vendor's spreadsheet program. Columns become
variables. Column heads become variable names.) or any other standard
Macintosh formats. (False. The standard Mac format for data exchange is
tab-delimited ASCII files and clipboard. Data Desk deals with both.)"
Frankly, I refuse to keep this up for 8 pages. It is not my obligation to
get StatSoft's fact straight. It is *their* obligation. Others who have
published comparative reviews (including other stat packages) have checked
with us (and, I presume, with other package vendors as well) to be sure
they had their facts straight. That would be a responsible thing to do.
> in one of
> your previous posts, you commented on SYSTAT's pointing out of Statistica
> bugs, and characterized StatSoft's quite similar response as "scurrilous"
> and indicated that they only found a few minor errors.
Actually, that's not the point I made. I noted that the bugs mentioned by
Systat included fundamental errors in statistics (such as erroneous
residuals and estimates of inestimable quantities), while the bugs found by
StatSoft were mostly obscure rounding errors found at the limits of
computational precision.
I hope that some of this is helpful to the net. I fear that it grows
tedious. I hope that StatSoft will elect to stop trying to make their
product look good by cutting down the competition. Such methods are
unprofessional.
-- Paul Velleman