Newsgroups: rec.games.int-fiction
Path: news.duke.edu!newsgate.duke.edu!nntp-out.monmouth.com!newspeer.monmouth.com!news.maxwell.syr.edu!newsxfer.interpacket.net!uunet!lax.uu.net!sac.uu.net!nyc.uu.net!world!buzzard
From: buzzard@world.std.com (Sean T Barrett)
Subject: Re: [Comp00] Size/Score Ratio
Message-ID: <G4A5B3.LEw@world.std.com>
Date: Sun, 19 Nov 2000 16:18:39 GMT
References: <20001118073926.28852.00000383@ng-cl1.aol.com>
Organization: The World Public Access UNIX, Brookline, MA
Lines: 123
Xref: news.duke.edu rec.games.int-fiction:58729

LucFrench <lucfrench@aol.com> wrote:
>Because I'm bored, here's a list of the size/score ratio (found by dividing
>size by score, and then rounding down) for all the Inform entries this year:

Everyone hates scatterplots!

Dividing comp scores by game size (or vice versa) and
examining the deviation from the mean is equivalent to
drawing a scatterplot, fitting a line through the origin,
and looking for outliers.  The scatterplot is potentially
more useful since it lets you fit a line that doesn't
go through the origin (well, you could do that with
least squares) or allow for curvature.  (In fact, a
basic rule I learned in college physics: never fit
a line to data without graphing it first.  If it doesn't
look like a line you shouldn't be fitting one to it.  For
example, the data could actually match two unrelated lines
if you're actually measuring two different phenomena.)

I've separated out Z5 games from Tads; the latter appear
after the Z5 discussion.  (For other categories of games
there aren't enough to meaningfully examine.)

= = =

                  Z5 games scores vs. size
score
8
7.5                                     *
7                           *    **    **   *  *         *
6.5                                    *   *   *
6                           *    *     S
5.5                             *      *    *
5                              **S     *
4.5                 *         *  S
4
3.5                      *
3                    *
2.5                *                 S
2                  *     D
1.5
1
0.5
0
       |         |         |         |         |         |         |
       0        50K       100K      150K      200K      250K      300K

Along the bottom are "Breaking the Code", "Asendent" (D), and
"Comp00ter Game" (S), suggesting that the payoff for "those kind
of games" grows less for increased effort compared to "normal" games.

However, it dawns on me that just as mixing Tads and Z-code games
together in one graph would be wrong, so it strikes me that comparing
Z-code games compiled with and without debugging, or with and without
strict mode, is equally fallacious.  (Indeed, in some sense, what you
might really what to compare is the size of the source code of each.)

So I went through and determined which Z-code games were compiled
which way; hence the 'S' and 'D' notation in the above graph (based
entirely on what the games reported in their banners).  So
I guess I'm wrong about the bottom outliers--both of which left
debugging modes in intentionally; they might fall back in line if
compiled without strict or debug mode.

It looks like a reasonably fit line to this data does pass
pretty close to the origin.  Although the "true" origin might better
be a game of 50K (you can't get smaller than that with Inform's
libraries) and a score of 1--and that point looks a bit below the
line.

But even so, I wouldn't recommend authors attempt to influence
their actions based on the assumption "bigger does better";
there's an awful lot of scatter: the seven games at 162K +- 5 scored
from 7.27 (2nd place) to 5.08 (31st place).  The eight games
scoring 7 +- 0.25 range from the extremes of "My Angel"'s 250K
to "Shade"'s 100K--subtracting out the roughly 50K default library
size puts them at a 4:1 size ratio versus a 1.05:1 score ratio.

= = =

                 TADS games score vs. size
8                                       *
7.5
7
6.5
6                             *
5.5                  *                             *
5            *  *
4.5          * *
4         *
3.5
3                   *
2.5
2
1.5
1
0.5
0
       |    |    |    |    |    |    |    |    |    |
       0K  100K 200K 300K 400K 500K 600K 700K 800K 900K

(Futz Mutz was omitted, since it has embedded music/art resources.
I also had to manually resize this graph when I added the rightmost
entry, so the data points may be off by +- 1.)

This looks like a much stricter line than the z5 games, and also
a line that doesn't go anywhere near the origin.  The two outliers
below the main line are "Jarod's Journey" and "Unnkulia X".

= = =

Future authors should also remember that the size of a game does
not merely reflect the quantity of the game, but often the quality
(the level of detail) of it as well; don't take this data as a
recommendation to make a big game in the sense of geography
or number of objects.  (In fact, I'd bet the aspect of the game
that's probably most closely correlated to the size of the data
file isn't the number of rooms or objects or puzzles, but simply
the amount of text the author wrote.  But then that could sound
like a recommendation for long-windedness, and that's certainly
not how you do well.)

SeanB
