The Two Year Effect: Confirmed?

Monday, December 28, 2009 Monday, December 28, 2009

A few weeks ago I described a theory I had regarding SL population that I called “The Two Year Effect.” The theory went something like this:

Without a defined requirement to stay, people tend to maintain interest in volunteer or hobby activities for a length of time between 18 to 30 months. About two years. If that’s so, then one reason for the apparent declining SL population is that the large bulk of folks who showed up in 2007 during the SL media hysteria are now expiring.

It was merely a theory, and I had no data to back it up.

Until now.

It turns out that reader Louis Platini took an interest in the theory and actually happened to have suitable data available for analysis. If you don’t know Louis, he’s the owner of Metaverse Business, makers of a wide variety of statistical counter devices for use by in-world businesses. His products range from free traffic counters to highly sophisticated systems capable of handling multiple regions and delivering far more than simple user counts. Do check them out!

The data collected by all these machines is safely stored by Metaverse Business so that their clients can make enquiries of their own statistics.

But Louis can also analyze this stored data, and he decided to do so after reading my article on the Two Year Effect. His results are very interesting, and you can read them in detail right here. The data should provide reasonably accurate results, as Metaverse Business has data on over 1.2M unique avatars. Louis employed data collected over all of 2008 and most of 2009 for the analysis.

The analysis attempted to determine the “lifetime” of avatars. In other words, the number of days between their first and last appearance. Some avatars “lived” only briefly, less than 100 days, whereas others have existed for many years. Louis then graphed the result, shown here:



The data shows an extremely steep departure rate right from the start. It appears that many people try SL for a few months and then depart (Give up? When is that easy-to-use viewer version 2.0 coming out anyway?)

Louis shows that the departures seem to follow an exponential rate for the first 400 days or so, then slow a relatively linear rate for the next 1000 days. After that departures slow right down to a trickle. Be sure to read his analysis where this is demonstrated with several explanatory graphics.


So where does this leave my Two Year Effect theory? It seems to be both wrong and right. Consider the exaggerated theoretical Two Year Effect on the chart above superimposed onto Louis’ actual data.

It’s wrong because it turns out that people are constantly leaving, not just after two years. The younger they are, the faster they tend to leave.

It’s right because almost everyone is gone after two years, and the large mass of 2007 signups must indeed be leaving (or already gone).

The fact that avatar lifetime is effectively only 2-3 years must be of grave concern to Linden Lab. A business that has temporary customers must have a strategy to get new ones to replace those departing, and that must be why Linden Lab is so focused on the experience of new residents.

12 comments:

Brent said...

One problem with this analysis: alts. How many people do you know who've been in SL for a few years, but in 2-3 different accounts over that time? Most Second Life users I know fall into that category.

Botgirl Questi said...

Very interesting!

I guess my first question is whether there is anything special about the avatars that were counted...for instance are the counters located at mostly certain types of business that would tend to exaggerate particular subcultures or interests, or exclude others.

It would be great to see log-in frequency as it related to avatar life. Are the people who leave not using Second Life very much, or do they maybe spike a few times and then burn out?

As you mentioned, seeing how avatar life relates to interest or subculture would shed more light on what's behind the overall numbers. I agree that creators probably have longer avatar careers. Also merchants and other business-related activities.

Ari Blackthorne™ said...

Very good theory.

I've always thought more or less the same, except I am of the belief there is a lot more recycling going on that anyone, including Linden Lab give credit for.

I've opined about this some time ago, curious as to your thoughts on this paradigm idea:

http://commonsensible.net/2009/08/second-life-endorphin-rush-drug-fix.html

annotoole said...

This data set is better described as "The Kingdon Effect since it only covers the amount of time Mark Kingdon has been involved.

But this is just a sampling. The data collectors have no way of knowing if an avatar account was canceled or simply stopped going to places where the data collectors are located which certainly do not represent the average business. Most likely people simply stopped patronizing expensive stores as soon as they learned their way around SL. Which, ironically, meshes with another recent post of yours.

Kwame Oh said...

Must agree with both of you with regard the data,having written somewhere before about people who flee the real world in attempt to burn bridges, and then proceed to slash and burn every river crossing they build in virtual world, thus changing avatars like you and I would change socks.

Factor in the prevalence of camping as a income generator for some during the period in question and the change in traffic calculation for search changes and many an avatar was made redundant.

Having said all that data of this sort is very interesting in the overall planning by any party taking part in the growth of virtual world on the whole.

I personally would love to see data collected, with candid answers from users/past users with regard trends over the years.

thanks for posting this as might cause some to dig deeper.

Julius Sowu Virtually-Linked London

Tateru Nino said...

@Brent: Only one person. Everyone else I know remains using their main.

Anonymous said...

WARNING Complex systems geekery follows :-)

I find you theory and the data analysis interesting. I would suggest however, that the curve is better described as a power law curve (see http://en.wikipedia.org/wiki/Power_law). Power laws are being identified in surprising numbers and wide range of disciplines leading some scholars to argue that the "standard" bell curve is in fact a special case, that the power law curve is in fact more common/standard.

\complex systems geek off

Angela said...

No offense to you, but using data that is collected from a spattering of sims, with no controls for confounds is not exactly what I would consider meaningful. Realistically, what is the r val? The mean? Standard deviation? You know anything that might even remotely indicate a statistically significant correlation wrt avie longevity?

Ken Taylor said...

There is doubt about the avatar lifetime analysis and I'd like to express more doubt. The data purports to measure retention in Second Life but it actually measures life of avatar from creation to the last time it was seen at a site monitored by Metaverse Business where it was seen more than once. This is not the same thing as the number of sites monitored is too small to be confident that an avatar that is not seen is dead and while the sample size is large being 732,834 avatars it is a small percentage of all avatars confirming the view that each avatar is not well measured. For example 238,012 avatars were only see in the first 0-100 days of life but the average age of first contact was 81 from http://www.metaverse-business.com/blog/avatarage.php .

It is also not a power law curve as suggested by anonymous as it is not scale invariant.

An interesting analysis for which Metaverse has the data is the distribution of the number of interactions with particular objects or distribution of the number of visits by avatars to a site. I suspect this might follow a weibull distribution http://en.wikipedia.org/wiki/Weibull_distribution with a shape parameter less than 1. See also http://urremote.com/index.php?title=WebTelerobotic_A_new_method_for_comparing_telerobots_for_operator_preference for other examples where weibull applies.

Gwyneth Llewelyn said...

This is quite interesting :) However, I hope it's not one of those correlation-is-not-causation effects. Still, as a valid theory, it certainly makes a lot of sense, and would neatly explain the sudden drop...

The question that begs to be asked is, "what can we do to revert the change?" Getting more people to join is naturally one choice — but actually the number of new registrations per day hasn't dropped so dramatically. People still don't stay around, they leave immediately after registering. The other possibility is to make SL compelling enough, in terms of stability and new features, to keep the interest — technological innovation drives new products, attracts new content creators with brand new ideas that are suddenly possible to be implemented, and brings more content for sale, and so on...

But I'm sure that's just the tip of the iceberg of the many things that can be done...

Botgirl Questi said...

I agree with Gwyneth that the primary problem is that the vast majority of new registrations don't make it past their first 90 days. I just posted on my proposal for a fix.

Nexii Malthus said...

You'll find that there is another strange phenomenon that is surprisingly common, although I have no hard facts to back it up how common it really is. When an abandoned account is revived by it's owner, the period of inactivity can range from 2 months up to 2 years.

Between residents, this has evolved into a memetic that surely everyone should recognize, be familiar with or have suffered from: Burnout.

Related Posts with Thumbnails