Forum Moderators: martinibuster
Mixed results have been found by different people. Some positive some more positive, some negative.
I'm not certain if Smart pricing is at the source of the price variation.
Here's how I think channels work.
-------------------------------------------------------
(the folowing is just my opinion and my opinion only)
-------------------------------------------------------
By creating a new channel, you are effectively creating a new artificial intelligence (AI) brain if you like.
A new AI brain needs training so it learns from impressions. If a channel is very specific. The AI brain will be trained to give adds which are very effective - yielding good results for some.
Deleting a good result channel effectively means you are killing the history of this AI brain.
If an AI brain has been trained in a wide range of topics etc. Results may not be relevant as it has learned too many things. Effectively making your targetting best for the whole site but not that good for a narrow segment. (an old saying Jack of all trades, Master of none) meaning that if you can do too many things, you can't specialise in one area. Killing this AI brain would effectively reset the history and attempt to re-learn what works best for the site. Like re learning to speak after a stroke.
It may be that ancient data on those channels was completely out of date thus making the entire site simply not work that good.
In saying that there's a flaw in the algo is in my opinion true as old data should be scrapped (forgotten) so only the latest data would affect targetting.
That's my opinion and I'm running with it!
Alain Pelletier
Mike
I deleted my channels with Anns experiment. I find it did help some.
In the past I had my various ad placements in channels. If my eCPM went down too much. I'd change my ad just so slightly as in changing the title color just a tad. I've posted several times that doing that for some strange reason made my eCPM go back up. I thought it was very strange and it made absolutely no sense that changing my link color a tiny little bit would lead to large increases in eCPM. Well, with your Artificial Intelligence idea, it now makes complete sense! It was not changing the link color that increased my eCPM, it was because I made a new ad with a new channel.
Thanks for the great idea, it makes perfect sense to me!
Interesting theory and it makes a lot of sense. You may have hit on something there. :)
Basically that was what I was doing but in changing the ad code (new ad) it also killed the incoming data that could have reached the AI even though the channel was deleted there was no way of knowing if the "trained intelligence" was gone without completely blinding the bot to that channel.
Good thinking.
Ann
Does Google use AI?
I can't see getting them to do what they do WITHOUT using AI.
What is AI? - wikipedia says:
Artificial intelligence (AI) is a branch of computer science that deals with intelligent behavior, learning and adaptation in machines. Research in AI is concerned with producing machines to automate tasks requiring intelligent behavior. Examples include control, planning and scheduling, the ability to answer diagnostic and consumer questions, handwriting, speech, and facial recognition. As such, it has become an engineering discipline, focused on providing solutions to real life problems. AI systems are now in routine use in economics, medicine, engineering and the military, as well as being built into many common home computer software applications, traditional strategy games like computer chess and other video games.
Are (sipders) bots intelligent?
Probably not they just visit and gather info.
Also, I have been using channels for the location of my banners (testing the size, and position) Perhaps that's not how they were MEANT to be used. Bare in mind that only the code designers know what they were meant to do so... instead of telling us they made us guess.
I think they were meant to be used category wise instead of site wise such as the use that I have right now to find out the time of day usage across 4 different web sites completely unrelated.
Ann
Basically AI works with neural nodes and the weight of those nodes is adjusted dynamically (trial & error) or statistically.
It is my opinion that they do not need to use the bots to target a new page. Using the bots just adds another degree of precision to the equation.
Deleting a channel would delete the weights of the neural net associated with that channel.
oh... oh... I think I'm starting to loose my audience.
Alain
[edited by: Khensu at 9:31 pm (utc) on Aug. 7, 2006]
Sorry if I missed this info elsewhere, but are you still channel-free now?
I'm new (2 months) and tried it but panicked because my earnings immediately tanked to 1-2 cents a click for a few days so I created new channels and things perked up again.
I'm interested if you are still cruising along channel-free? I really appreciate your insights here.
panicked because my earnings immediately tanked to 1-2 cents a click for a few days
It sure would be nice to know whether each person posting about positive or negative changes has actually done the mild calculations required to know whether those changes have much significance compared to their normal random variation.
How would one do these calcuations? I'm game but need direction.
If we say, for example's sake, my epc shifts between 80 and 120, I was suddenly getting 40-60. Like a light had been switched off. No other changes on site.
It may provide all of us an insight as to how to better use the channels or not use them to our advantage.
Simply put, It is my impression that it's a tool that google has provide us but has forgotten to tell how to use it.
If it was strictly for tracking purposes as google claims, deleting a channel should have NO effect. Since some see great effect, I am suggesting that there is a mechanism at play and it is I believe it is AI.
Alain
How would one do these calcuations? I'm game but need direction.
Crude, simple, but much better than eyeballing:
You probably have some kind of spreadsheet software laying around that is happy to calculate variance and related numbers for you. Just load the variable of interest (daily revenue, average daily CPC, whatever) into the spreadsheet (e.g., the last 90 days worth), and ask it to calculate the mean and standard deviation (the software help file should tell you how). Don't include the recent days that are the result of testing a new change.
We can make long and tedious arguments about when and why this measure might be misleading, but the point is that it's a good start, and much better than nothing.
Once you've got a mean and a standard deviation, then you can strongly suspect that any number that lies more than 2 standard deviations from the mean is significant. Example, if your mean CPC is $.10, and the standard deviation is $.02, then seeing that Ann's change is bringing you daily averages of $.13 CPC is not highly compelling evidence, but $.15 would sure look interesting.
Also, if your mean is $.10 and your standard deviation is $.07, you know that you've got a heck of a lot of variation. Maybe you just have too little traffic to do much good testing of changes at all. If you've got a whole lot of variation even with a sample of 90 days, well, you may just not be able to do a good job of isolating the effects of any change to your website.
You might be able to "fix" this excessive variation by selecting a bigger "bucket size" than "day". For example, if you take the average weekly CPC, and examine the last 30 weeks, you might see a lot less variation. Unfortunately, having to look farther in the past can expand the effect of confounding factors: new content, Google algorithm changes, adding and losing advertisers, etc. The smaller your website traffic, the harder it's going to be to correctly assess the effect of any changes.
There are some other obvious factors to control for. Did your traffic volume change significantly during the test period? Then that might be the real source of a significant change in CPC, total revenue, etc.
Likewise, did your traffic sources change significantly during the test period? If you dropped in ranking for search term X but raised your ranking for term Y, you might end up with roughly the same overall volume, but a significantly different kind of visitor, with different product interests and clicking habits, which might also be the real source in a significant change in CPC, total revenue, etc.
If you don't already know roughly where your traffic is coming from (e.g., the search term categories or website links that account for 80% of your traffic), then stop worrying about your earnings and focus on understanding your websites basic numbers. You don't want to be one of those folks that posts "Google took away half my traffic!" and then just has to say "I dunno" when someone asks "well, where was the traffic coming from before?"
Finally, the other obvious factor you want to have a handle on is your advertisers. One high-paying advertisor leaving can easily skew the results for modest-sized websites. Google does not make this easy for you to track, but you should already have made some efforts to identify which pages are providing most of your revenue. You can use the AdSense Preview tool for your top N URLs, snap a screen shot, and then compare later to see whether you should consider an advertiser shakeup as the source of some other change in your money stats.
Example: I see my results so far in August look awfully low compared to my July results. Is Google stealing money from me because I had a good month in July? Did SmartPricing suddenly hammer me because I changed a single paragraph on the home page? I download July into an Excel spreadsheet. I make a cell called "=AVERAGE(F2:F32)" (column "F" happens to have my daily earnings, while rows 2 through 32 contain each day of the month). I make another cell called "=STDEV(F2:F32)" to get the standard deviation.
I see that my mean daily revenue for July was X, but my standard deviation was more than 50% of the mean. Yikes! Suddenly the fact that August daily revenues are so far about half of the July average does not look significant at all. My conspiracy theories are dampened, and I am reminded that, above all, I need larger amounts of qualified traffic at the moment.
Do you see how people posting things like "I changed X and my revenue doubled!" can be completely useless information? Likewise, "I changed X and my revenue was cut in half!" can be completely normal random variation for many websites, and have no relation to the change being tested at all.
For every management variable that is difficult to measure, there is some measurement that can be done that is better than nothing.
for example's sake, my epc shifts between 80 and 120, I was suddenly getting 40-60
And, for example's sake, what was the mean and standard deviation of the previous 90 days worth of average daily epc?
which means my mean of c.50 when I dropped channels was significant.
For every management variable that is difficult to measure, there is some measurement that can be done that is better than nothing.
Great message, Ron, a good coverage of basic stats and inference that isn't intimidating for people with no background.
It's actually amazing what you can do and learn with statistics, and I wish I had the time and energy to play with number re: both adsense and google.
For example, what's the correlation between number of adwords clicks and adsense income?
When does smartpricing kick in?
Tons of stuff, and if you get into statistical tests it gets more interesting. It does take time, interest and skill, and a decent statistical analysis package.
Anyway, good post.
It probably won't help much with the conclusion jumping, but maybe for some...
rough and ready:
mean: 90
SD: 14which means my mean of c.50 when I dropped channels was significant.
You also need to calculate the SD for the post change to eyeball.
It can help you figure out what's going on.
