Google Website Optimizer Case Study: Daily Burn, 20%+ Improvement | Hacker News Reader

Google Website Optimizer Case Study: Daily Burn, 20%+ Improvement

fourhourworkweek.com

51 pointsdivia22 comments

"Conclusion: Simplified design improved conversion by an average of 20.45%." is the most important sentence in this post.

To be honest you don't have to use Google Website Optimizer to notice that.

It is easy to believe the things you want to believe. A/B testing is useful for convincing you of the things you don't want to believe.

Example: I'm releasing my Rails A/B testing framework this Sunday, with a case study writeup taken from my site. My site currently allows people to defer signup for the trial by means of a guest login. That sort of option is popular here. I'm told it increases usability and user engagement, right?

Without ruining the surprise: that is a testable hypothesis. ;)

A/B testing is great, no doubt. The lesson of solely simplifying is just not a great outcome of such a test. It's like driving a car for the first time and then telling everybody that it's faster than walking.

Yeah you do. I ran a test on a friend's site, convinced that a short version would outperform the long-copy version. Conversions dropped by about 30%, which would have cost him tens of thousands of dollars per year.

There is some "New Math" being invoked here. If the conversion rate goes from 24.4% to 29.6%, it's disingenuous to claim a 21.2% improvement.

I stopped at the first misuse of statistics, but by that point, they had taken out informative links from their web site and turned it into a landing page. Frankly, a 5% change in conversion rate is in the noise compared to the effect of targeted advertising (and I have no faith that these sloppy reporters actually did a properly controlled experiment).

Edit: Downmodded for asking for real math over marketing math?

It's been a while since I've taken a math class, but I'm not sure what the problem is with calling it a 21.2% improvement.

The old conversion rate was 24.4%. The new one was 29.6%.

(29.6 - 24.4) / 24.4 = .213 = 21.3%.

Is it misleading because their number of conversions is so low? (Not trying to be snarky, I have no idea)

The improvement in the conversion rate is (29.6 - 24.4) = 5.2%.

Reporting "a percent of a percent" is downright misleading because it is entirely dependent on the starting baseline and also because conversion rates have high natural variance. Imagine an improvement from 2% conversion (for a truly awful site) to 2.6% (for a site just as awful). That "30% improvement" just isn't. If they were intellectually honest and called it a "0.6% improvement," anyone would be able to see that the claimed improvement is well within stddev.

But the underlying metric that actually matters, the number of users converted, did actually improve by 21.1%.

I think it's just a matter of using confusing labels to refer to the percentages/improvements.

In other words, you don't really care about the second derivative (the rate at which the conversion rate) is improving as much as you care about the initial rate.

anyone would be able to see that the claimed improvement is well within stddev.

You appear to know words associated with statistics, but treat them as if they were magic incantations utterly disassociated from actual math.

I'll bite: what is the standard deviation in that example. I'm looking for a two part answer: a) a number, b) what the number is measuring.

Let me be intellectually honest with you: there is no basis for assuming that an improvement from 2.0% conversion to 2.6% conversion is not statistically significant.

You need sample sizes to even attempt to do the math. For example, try doing a chi-squared test on 100,000 people converting at 2% versus 5,000 converting at 2.6%.

I'm thinking you'll find that you reject the "just as awful" hypothesis with over 99% certainty.

>what is the standard deviation in that example...You need sample sizes to even attempt to do the math.

First, you're asking the wrong person to supply the missing data. The onus is on the OP to justify their stats, not on the person calling BS on the misuse of statistics.

Given that the original article is missing the critical numbers and presenting the rest in the most self-serving manner possible, the only data anyone has to go by is their own experience with conversion rates. We don't even know what "conversion" means for this website because they don't define it, but that justifies making reasonable assumptions (e.g. users who land on page A end up on page B, for a website in some industry, involving some side-effects to get from A to B that are left undefined by the OP). In my experience, the behavior of users on a web site is quite fickle and dependent on the surrounding advertising campaigns, as well as the weather, the time of day, and the timing of holidays. And maybe a billion other confounding factors. So, is the 5% improvement seen here significant? We don't know, there's no evidence to assume it is, and that's precisely my point.

Huh.

The number that matters is "how many more users are signing up since we changed"? The answer is NOT 5.2%.

Example: I have a 1% conversion rate on 100,000 visitors. 1,000 customers! Yay! I change something and I get it up to 2% on the next 100,000 visitors. 2,000 customers.

Which is a better description: "I doubled my conversion rate" or "I increased my conversion rate by 1%?" I'd go with the former.

An equally valid question to ask is: "How many users are bouncing away from that website after landing on the first page?" That raw rate improved by 5.2%. When you quote percentages of percentages, the rates vary widely depending on your point of view -- in the original example, an optimist might report a 20% improvement, looking at the users who reached point B, and a pessimist might report a 10% improvement, looking at the users who left without reaching point B. This is only partly why it's a bad idea to quote percentages of percentages -- they are dependent on the baseline. There are at least two baselines to choose from (successful conversions vs bounces). Also, percents of percents hide the task difficulty, which varies depending on the starting point (going from 1% to 2% is probably not hard, going from 99 to 100% is probably very very hard).

But the other reason why it's a bad idea to quote percentages of percentages is that there is no indication whether the claimed improvements are actually statistically significant. I didn't see any data in the article to indicate they are, and what little I know about conversions from my own observations is that they vary by at least 10% depending on totally random factors.

>Which is a better description: "I doubled my conversion rate" or "I increased my conversion rate by 1%?" I'd go with the former.

I see your point and I'd use it if I needed to flatter myself. But I'm sticking to the latter. And I'd be sure to mention the starting baseline as well.

"I see your point and I'd use it if I needed to flatter myself. But I'm sticking to the latter. And I'd be sure to mention the starting baseline as well. But I'm sticking to the latter."

Help yourself. Just don't use that type of language in any conversation with anyone else on the planet because you'll likely confuse the hell out of them (take note of the "wtf is this guy talking about" downvotes you've received on this thread)

It's not about flattery. It's about metrics that matter and revenue. Doubling your conversion rate pretty much doubles your revenue. That aside, the goal here is communication. I'd wager that if you said that you doubled your conversion rate, people would grok what it meant (1% to 2% or 5% to 10%-- either way, a big win because it'd double revenue and profit down the funnel). If you said you increased your conversion rate by 10%, I'd wager people would assume that meant something like 10% to 11%.

>take note of the "wtf is this guy talking about" downvotes you've received on this thread

I don't rely on opinion-polling to determine who is right, esp. when it comes to questions related to science.

>conversation with anyone else on the planet

On your planet, a move from 0.0001% to 0.001% would be reported as "an improvement of 1000%" No thanks.

> It's about metrics that matter and revenue.

Yes, we established long ago that it's about marketing-speak versus math.

So, since you have so many upmods, perhaps you can tell me what happens in your world when the conversion rate goes from 0% to 1%? :-)

You're saying that someone who talks about the percentage growth in their customers would be just as concerned about the percentage shrinkage of their non-customers? Unless you're Facebook or Coca-Cola, you probably don't assume that the majority of people in the world will at some point use your product. So it doesn't make sense to measure your shrinkage of non-marketshare.

The percentage change in users is the interesting number. If someone said "My conversions went up by 100 per week!" I would say "What percentage increase is that?" If someone said "My conversions went up five percentage points!" I would say "What percentage increase was that?" If they said "My conversions went up 20%," I would know what I wanted to know.

Values reported as percentages should generally be values between 0 and 100. So when the real conversion rate goes from 0.002% to 10%, the reported number should not be 499900%.

I expect hackers to use metrics that are at least defined over the range they are supposed to be used. Imagine that the conversion rate goes from 0% to 1%. I report 1% improvement, the value I measured. You? (1-0)/0 = what percent?

It's _not_ a one percent improvement (1%). That is mathematically incorrect.

It's a one percentage point improvement. Wrong terminology is responsible for 95% of this thread.

No, because that means you're using different metrics to compare the same changes, e.g. "Over the last five years, one of my stocks went up 75%. One went up 100%. One went up 125%." What purpose is served by making the last two numbers a different scale? (If you'd like to stick with the percentage terms, pretend I'm talking about dividend yield and not stock price)

Bad example. The whole discussion is about reporting percentages of percentages.

Look, it's pretty clear that hackers have left the building and let marketroids rush in. Feel free to use your ill-defined metric. If you have the least bit of intellectual curiosity, you'd let everyone know what you'd report when the value goes from 0 to a non-zero value. That is all.

Yes, the whole discussion is about that. Apparently only marketroids are able to handle multiple layers of abstraction -- yes, you can take a percentage of percentages, just like you can take a percentage of dollars or a percentage of conversions.

"I increased my conversion rate by 1%?"

It wouldn't be an increase of 1%. It would be an increase of one percentage point OR an increase of 100%.

This entire thread could be resolved by acknowledging the differences between percent and percentage points.

There's a difference between a 20 percent increase and a 20 percentage point increase.

Reply on news.ycombinator.com