Calendar
Network
Feeds
- 48
- Sphinn It!
Went Hot: May 14, 2008 - 5:05 pm
Posted By: Lyndon 72 days ago
Topic Type: News Story (Jump to http://www.searchlaboratory.com)
Category: Google AdWords
14 Comments
Went Hot: May 14, 2008 - 5:05 pm
Posted By: Lyndon 72 days ago
Topic Type: News Story (Jump to http://www.searchlaboratory.com)
Category: Google AdWords
14 Comments
Top 10 Most Sphunn
In What's New
- How Net Promoter Score (NPS) Can Di... (21)
- On-Site Geo-Targeting and Local Sea... (20)
- Misinformation In the Blogosphere (16)
- Google's Knol Launches: Like Wikipe... (15)
- Negative Keyword Research Tools & T... (15)
- SEO Consultation Ruining Industry R... (15)
- Propeller 2.0 Launches: Ditching Th... (12)
- “Scads:” The Next PPC Battleground?... (11)
- Slagbait - The Future for SEO Blogg... (11)
- Hollywood's Behind the Internet Tim... (11)
Latest Comments
- We had it set so if you got logged out of your stats it defaulted to BL…
- Halfdeck,My 20 years old daughter would be very mad at you if y…
- "I notice Lyndon has also blogged the post too…
- "Oh...to work for a company with the balls to turn a job down! :) IF ON…
- Btw, this thread and one of my yesterday t…
- Alternate headline would be"The day seo ate itself"…
- True you have to learn how to fire clients who are too demanding on your ti…
- "Oh...to work for a company with the balls to turn a job down!"…
- - read more -
Save the date for:
• SMX Local & Mobile - San Francisco, CA (July 24-25) See the agenda, and register now!
• SMX Sao Paolo - Brazil - (Aug. 7-8)
• SMX China - September 23 & 24, 2008
• SMX Stockholm - September 23 & 24, 2008
• SMX East - NYC - (Oct. 6-8) Registration is now open.
• SMX London - November 4 & 5, 2008
Comments
That's some pretty cool math. Or should I say "those are some pretty cool maths"? ;-) Yeah, I know, my boss is English too. Anyhow, this is a not too intimidating explanation... I wish there were an online tool for calculating this like splittester.com- is there one?
I hate to rain on the party but isn't this just a sales page? It's not really a guide on how to do the maths even though it does cover some of the points... Maybe I'm just in a bad mood today :-)
@bbcarter - Try this tool which is free: http://www.vertster.com/adwords-tool/default.asp
We're releasing a tool which will run over your whole account automatically and notify you when you need to re-write ads. Drop me an email or follow our blog if you want to know when it's being released!
@ Tom
I personally think this page is of value ... ok so the bottom bit advertises his services ? so what ? Ok so its not the best written article I have ever read but it is a deep subject that is hard to explain.. Advanced maths probibilty ... How do you get your head around that..? I think they are trying to explain that if you use maths instead of guess work then your PPC account will take a turn for the better.. When I personally read the page I didnt think they were trying to educate me on how to apply the math.. I think they were just saying you should be and heres why ...
You actually say in your comment above that it is a sales page then drop this line in follow our blog ... Giving yourself a bit of self promo
What kinda algo or statistics is your tool going to use to tell you when to update your ads
In Tom's defense, the link isn't to his own site...
sales pitch at the bottom, but it's also an informative article so by the time I got down there, I didn't mind. THIS is how you do it.
Hey ciaran.. I never mentioned the link :) I was just pointing out the irony of "sales page" then the "follow my blog " :)
I'm not so sure I trust the 'maths' these folks are doing. I'd have to hit the textbooks again to confirm, but some of it sounds wrong.
For example, most political polls use a sample size of 1000 people, which gives something like (assuming a normal distribution and random sampling of the population) results accurate + or - 2.5%, 95 percent of the time. in other words, almost all the time with a 1000 samples your results will very closely resemble the entire population.
Yet they discuss a sample size of 2000 like they're unsure if that's big enough statistically. And then they drop in some mention of 'if you get one more click shortly then that would take you from profitable to unprofitable.' Which leads me to question whether or not they should be using a normal distribution to model that data set. Gosh, I've forgotten the distribution for that...what's the one where they used it to model the deaths from kicks to the head by horses in the French Army? Not the normal. Pareto? No, that's not it either.
I've seen companies who had software where you put up nine ads with slight variations and they would statistically analyze which ad combination would work the best - and that stuff was bang on. So stats is a good thing with PPC. But either the folks writing this don't know as much statistics as they want to portray, or they've written a poor article (or, I've forgotten so much of my stats that I'm full of crap. That's possible too).
Hi, I'm Ian and I wrote the article! Just to explain the math(s) a bit further, and to answer Wheel, the sample size of 2000 you mention is in case 2. The 2000 is the number of impressions. In this instance I am actually using the sample size of 40 because this is the number of clicks. When we are valuing a keyword in this way we care most about it's conversion rate (clicks to conversions) not its click through rate. This is why you are seeing it as a large enough sample and I am not.
Also, you quote me saying 'If we get one more click shortly then we move from unprofitable to profitable'. The article actually says if we get one more conversion shortly because it is conversion rate that we are really interested in.
The statistical significance in a situation like this is dependent on the sample size, but also our expected and actual results. For example if we are tossing a coin (which we expect to be 50:50 heads and tails) and we toss 10 times and get zero heads. This is a lot more significant than if we have a keyword with an expected conversion rate of 10% and we get 10 clicks and zero conversions.
You are right that this is just like political polls. I don't think you are full of crap but I think I have confused the matter by putting in the number of impressions in each sample.
I think the maths is all correct, however the point of the article was more to explain the issues you face and that most people just guess. If you guess at least you make decisions. If you wait for the stats then you do nothing most of the time on your deep keywords when there probably are other clues you can use (by clever grouping). We employ Maths Grads for PPC account management because it is that hard to maximize.
Sales page? Well I wrote it to explain this particular element of PPC management and how hard it is. We do this for a living so it's hard to not have it linked in that way.
I was considering putting a spreadsheet on the page as a download so that you can plug your own figures in and see if they are significant. I think that would be more useful than the Vertster tool because you have to do one at a time with that web based tool. Let me know if anyone would find that useful.
please send me or post the worksheet. my numbers dont add up the same as youres, and its been too long since stats class. :)
hey, this is nice calculator:
Sample Size Estimation Analysis:
Q: How large of a sample size do I need?http://www.websharedesign.com/sample-size-estimator.php
no its not my site
>>>>Also, you quote me saying 'If we get one more click shortly then we move from unprofitable to profitable'. The article actually says if we get one more conversion shortly because it is conversion rate that we are really interested in.
lol. That just proved that in fact you don't know anything about statistics. It makes absolutely no difference WHAT you're measuring, either you've measured enough to give you a small enough confidence interval to make a decision, or you haven't.
But you're not doing that. If waiting for one more event is going to drastically change you're results then you're using the same kind of 'statistics' as people who think that 99 tails means the next one is more likely to be tails.
Until you've actually got some verifiable postings that show you actually know statistics, you're bandying about like you're using 'maths' to prove your stuff just makes you a bullshit artist.
Desphinn cause that's what's going on here.
You are not understanding the point I am making in this particular bit. I am saying that 40 clicks and 1 conversion is a very likely result from a keyword with an underlying conversion rate of 5% (i.e. on average it would get 2 conversions in every 40 clicks). My point is that in this case (case 2) we have NOT got enough data to make this decision.
I think you need to read it more carefully. You are attacking it (and desphinning) when you don't understand it.
The maths in here is spot on. Maths is fact, so I am not going to get into an argument over the numbers. A major part of my first BSc was in Stats, for which I got a 1st so I know what I am doing and I am not making this up.
>>>> My point is that in this case (case 2) we have NOT got enough data to make this decision.
So prove it. Actually prove it.
I'm glad you've got a stats education. So do I.
In Case 2 we have 40 clicks and 1 conversion. Conversion rate of 0.025 or 2.5%.
Our break even in this case is 2 conversions; a conversion rate of 0.05 or 5%.
The statistical question we are asking here is:
"What are the chances of getting the results we saw (1 conversion in 40) if the underlying conversion rate is actually 5% (2 conversions in 40)."
Please tell me you don't need the sums to be able to say that for this case we are nowhere near 95% confident that this keyword could not be break even.
If you do need the sums, they are actually in the article, and they say that the chaces of getting this result if the underlying conversion rate is 5%, are 24%. This is high, so statistically we cannot condemn this keyword as bad on these figures.
Please let me know which bits of this you don't understand.