Statistics on impact of download size

[Thorbrian]

Hi All,
After reading the thread on download size:
http://forums.indiegamer.com/showthread.php?t=6244
I figured that there is some data on the impact of download size from Reflexive Arcade that may be worth sharing. So below are links to some graphs showing the average rates of various statistics for games of different download sizes, at MB resolution (so all data for a game has been averaged in with data from games with the same integer download size) taken from 3 months of data

This graph shows what percentage of initiated downloads are completed, averaged by download size:
http://reflexive.com/somethingelse/d...RateBySize.png

This graph shows what percentage of downloads happened at a speed at or below what I considered the max speed of a modem (53.3 kbps), averaged by download size:
http://reflexive.com/somethingelse/d...peedBySize.png

This graph shows the relative measure of sales, downloads and conversion rate, averaged by download size (note that each datapoint is the average for all games of a certain size, normalized to the max value for that statistic)
http://reflexive.com/somethingelse/d...InfoBySize.png

...All the charts look very noisy, I know. The reason why is basically that all the statistics measured vary quite a lot from game to game, far more than the statistics vary based on size - the noisiness of the data may put someone off of it at first, but I found signal to noise ratio for metrics by download size is a useful piece of information about download size, in and of itself.

[Sakura Games]

Wow so from those graphs seems that a game 20mb of size can do pretty well

[Phil Steinmeyer]

Excellent data - thanks for posting it!

[Thorbrian]

One other thing I forgot to post - for Reflexive, downloads that happen at or below dialup speed are about 40% more likely to result in a sale on average than those that complete faster than dialup speed. So on average, if 10% of a game's downloads were at dial-up speed, then about 13.5% of it's sales were from dial-up speed downloads.

[soniCron]

Interesting. I'd chalk that up to the number of downloads a broadband user is capable of achieving, versus that of a dial-up user. Sound reasonable?

[Game Producer]

Nice stats, thanks for sharing.

Is there some explanation for the low sales of 50 meg games, but high sales for less than and greater than 50 meg games? Maybe the number of *good* 50 meg games is reletively low at Reflexive?

[cliffski]

don't forget half life 2 sold a lot of copies through download, as did GalCiv2. As I recall, both were in the gigabyte range.

[dmikesell]

Originally Posted by cliffski

don't forget half life 2 sold a lot of copies through download, as did GalCiv2. As I recall, both were in the gigabyte range.

If the game rocks, download size is irrelevant. Word of mouth leads to viral marketing leads to huge success.

[terin]

I'm curious what happens in the user psyche between 24 megs and 26 megs... apparently that 25 meg download has the mark of the devil (or there's something wrong with the normalization of this curve)

Seriously, what this data tells me, based on my interpritation is there is a slightly negative trend on download size -> completion ratio that reduces download success approximately 10% between 5 and 55 megs, so a decrease in download success of .2% per MB.

The KEY missing factor to this useful and interesting information is what is the DOWNLOAD PERCENT over this same graph and timeframe. How many users hit the information page vs. download.

There may be a percent of people who see the download size and don't even BEGIN the download, which would further exaserbate (probably) this .2% per MB trend. I'm really surprised by that .2% thing btw, based on eyesight alone it seems to be a VERY normal trend between 5 and 55, where there is basically a drop in completion every MB. If anyone feels up to it, I would like to see a statistical "trend" analysis between 5 and 55 on this graph so I can confirm my guess of a .2% per MB.

-Joe

[Thorbrian]

Whoops, I realize I didn't give you guys any info on how many games Reflexive has at the various download sizes...
http://reflexive.com/somethingelse/d...stribution.png

Originally Posted by Polycount Productions

Is there some explanation for the low sales of 50 meg games, but high sales for less than and greater than 50 meg games? Maybe the number of *good* 50 meg games is reletively low at Reflexive?

Well from the distribution of game sizes you can see that the data points above like 23MB or so are pretty much just single games here and there - so those points say much less about the average game at that size than the other points do.

[Thorbrian]

Originally Posted by terin

I'm curious what happens in the user psyche between 24 megs and 26 megs... apparently that 25 meg download has the mark of the devil (or there's something wrong with the normalization of this curve)

haha... well hopefully the game size distribution graph will help you understand the anomaly better

Originally Posted by terin

The KEY missing factor to this useful and interesting information is what is the DOWNLOAD PERCENT over this same graph and timeframe. How many users hit the information page vs. download.

I'm afraid we don't have any downloads per impression data right now - but if I end up getting some, I'll post it here.
... I could be wrong, but I do think however, you can get a sense of how many people are turned off by the download size by looking at the difference between the completion rate graph and the % dialup graph.

Originally Posted by terin

If anyone feels up to it, I would like to see a statistical "trend" analysis between 5 and 55 on this graph so I can confirm my guess of a .2% per MB.

Linear Regression sayes .21% per MB:
http://reflexive.com/somethingelse/d...Regression.png

I was a bit suprised to see the completion rate data come out that way myself... I kind of had to double check it before posting to make sure it wasn't random bad luck... even though that graph is for the 3 month window from April 24th, 2006, a similar trend is shown over the entire last year. Also, I should note that even though the statistics vary quite a lot from MB to MB, the individual game data that was averaged to make those data points was really quite consistent - meaning the reason the average values jump so much is that difference between individual games is so large, as opposed to it simply being too small a time window or anything like that...

[amaranth]

Those graphs are great! Thanks for sharing them with us.

[Grey Alien]

coool info thanks. It's also interestingthat dial-up customers are more likely to buy!

[Phil Steinmeyer]

Resurrecting this old thread...

I'm at the point of deciding on music for my next game, and thinking about it in relation to download size.

Right now, my game is about 4.2MB, compressed via InnoSetup, WITHOUT any music.

I doubt I'll add much more graphical content, though install size will almost certainly go up some at the portal level, with the addition of their DRM, logos, etc. I could probably squeeze of fatten the 4.2 a bit by tweaking compression levels on JPGs and such.

Right now, I'm using placeholder music that's OGG compressed straight music (i.e. not MOD, MIDI, or anything like that). I've got quite a bit there (~35-40 minutes, IIRC), to reduce repetition, though I've compressed it pretty heavily (22K stereo instead of 44K, and the max oggenc compression settings). At these settings, the music is 8.8 MB, bringing the overall install size to ~13.0MB

I haven't paid for music (using placeholder by permission of the musician).

I could:

1) Switch to MIDI or something else highly compressed and possibly squeeze under 5MB for the whole game install
2) Use a somewhat fatter format like compressed MOD and shoot for ~7MB
3) Do what I'm doing now - good sounding OGG music, 13-14MB installable
4) Fatten up my OGG compression settings (44K music), and roughly 21MB insstallable.

Unfortunately, the linked to graphs above are now gone, but from my memory of them, and the subsequent discussion above, the link between file size and various measures of success doesn't seem THAT strong. I'm inclined to stay on my current course (#3 above - 13-14 MB installer), but am interested in opinions, and especially data points, particularly from those running portals who can measure this stuff across multiple games.

Basically, I'm weighing the quality/size tradeoff, and am inclined to go for quality, but am looking for input.

[Phil Steinmeyer]

Also, I seem to remember that some portals used to have special icons, visible in the main game lists, to denote games below a certain size (5MB?). For a site that uses these, I'd imagine there'd be an extra pop in downloads for games small enough to get that icon.

My quick perusal of RealArcade, BigFish, and Yahoo failed to see such icons though. Are they a thing of the past, or do some sites still call out small download games?

[soniCron]

If it helps: A couple weeks ago, I checked the filesizes of all the top portal games, and not one was smaller than 15MB - several were over 20MB.

[Sillysoft]

Trygames.com has a category of 'Under 10 MB'. That's the only time I've seen such a division like that.

Personally I'd go with #3 in your options. I think 13 MB is pretty doable even for dial up users. Although you probably want to test out the different compressions on the final music to see if you can notice any difference, and find out where the sweet spot of size/quality is for the specific songs you have.

[Drake]

A possible option #5 - maintain or increase your bitrate, but mix down to mono. It depends on the actual music and your taste, but I generally prefer a cleanly compressed mono track over a stereo one with audible artifacts. I don't think most people even notice the lack of stereo, especially if it's just BG music and there are lots of SFX filling out the mix.

[Phil Steinmeyer]

When I was first playing with bitrates doing some tests a while back, I couldn't tell the difference between stereo and mono for the piece I was working with.

That's I noticed that my left speaker was dead. Had to buy a new set

You're right it depends on the music. For well crafted music though, I generally will take the stereo over the bitrate...

[MrMark]

i'm getting 404 errors on the graphs

[RinkuHero]

Me too. I think they're too old.

[Thorbrian]

Sorry about the dead links - I fixed them now.

[Artinum]

Originally Posted by Thorbrian

Whoops, I realize I didn't give you guys any info on how many games Reflexive has at the various download sizes...
http://reflexive.net/somethingelse/d...stribution.png

You missed one. 404 here.

[Bad Sector]

http://reflexive.net/somethingelse/d...stribution.png

This and some others still don't work :-D

[Thorbrian]

Ah, yes - thanks, forgot about follow on posts. I edited the posts so they should be fixed now (FYI, reflexive.net doesn't work for getting to those games - I had to change the links to reflexive.com)

[Musenik]

It would be great to see how these graphs changed since last year. Got any current data?

[Thorbrian]

Sure, here's updated graphs:
Completion rate by size
Dialup by size
Size Distribution
Performance by Size

Looks like games are only slightly bigger on average, download size discourages download completion in the same way it always has, there are less dialup users but still between 4 and 12%, but some big games have come along and been the best performers.

So looks to me like basically the same story, smaller is slightly better, but game quality itself easily trumps the effect size plays.

[Musenik]

Cool! Thanks.

[Game Producer]

Originally Posted by Thorbrian

Sure, here's updated graphs:
Completion rate by size
Dialup by size
Size Distribution
Performance by Size

Looks like games are only slightly bigger on average, download size discourages download completion in the same way it always has, there are less dialup users but still between 4 and 12%, but some big games have come along and been the best performers.

So looks to me like basically the same story, smaller is slightly better, but game quality itself easily trumps the effect size plays.

Those last pics were for 11/06 - I was wondering if you again you'd have more recent data. (Perhaps Jan - April 2007?)

Thanks for those earlier pics, I really appreciate this!

[Escapee]

Good Info, TQ