Latest Netcraft survey shows Apache increase

The latest Netscraft Survey is out. Apache enjoyed an over 1 percent increase, with Microsoft and Netscape showing some decreases. According to the survey, Apache has a 54.81 percent "market share." Also reported is the fact that Webjump actuals uses a hybrid setup with NT serving static content and the dynamic content with a Solaris/Apache/Perl system. Tucked away in the report is a small factoid that PHP is on over 1.1 million domains.

Comments

sounds like a monopoly

[maskatron]

maybe the government should break up apache into....err nevermind.

Re:Margin of error?

[King Babar]

I just have to wonder, with an "increase of 1%" what the margin of error is - does this represent a real increase?

Others have mentioned possible problems in interpreting such data which include (but are not limited to) the following:

1. This is a population, not a sample
2. Even if it were a sample, it wouldn't be a randomly selected sample.
3. Even if it were a randomly selected sample, it would be a random sample of domain names, not host machines.
4. And many, many, more. :-)

OK, having said that, it might be useful to pretend that none of these were concerns, and that we really did want to know whether a 1% increase in the number of domains served by Apache meant anything. Here's the short answer:

I can't tell you that.

This is especially true if the domains surveyed in some sense are the population. In that case, whether or not you care that Apache added 500,000 domain names to the population while IIS added 125,000 is basically up to you. There are many explanations for why this could have happened, and not all of them are very interesting. (Again, others have pointed out why.)

Personally, I would have been more interested in certain kinds of longitudinal breakdowns rather than the overall numbers. Some of those questions would include:

1. How many sites went from IIS to Apache vs. vice versa?
2. How many "new to the survey" sites went with one solution or another?
3. How many "new to the survey" sites are actually new to the web?
4. How many sites running Apache (or IIS) closed down during the last month?

Call me a geek, but these are questions I think could be more interesting to ask. And, yes, some of the answers to these are given or hinted at on the netcraft website.

But there is one more question, which is the one the original poster asked:

But what if this really were a sampling question; is a 1% difference likely to be reliable?

If all N of the netcraft domains were independently and randomly sampled from the total population of domains, then a 95% confidence interval for a given market share, M, where M is between 0 and 1 is:

[M - 1.96*(M*(1-M)/N)^.5, M + 1.96*(M*(1-M)/N)^.5]

For Apache's market share in November, we would get the interval [.5479, .5483]. For the October share, the interval is something like [.5365, .5369]. Those are pretty tight intervals, but the sample size is over 8 million...

And this is the real point: when you have random samples this huge, error bars are pretty danged small. So it's too bad these really aren't random samples...

Not 54, but 59.53%

[rw2]

An even more interesting page is here.

It shows the usage by platform. There are a couple significant Apache derivatives that aren't grouped into the more conservative number that is used for the graph.

In fact, the bulk of the tailing off shown in the graph for Apache was actually slack picked up by Apache derivatives!