Forum Moderators: martinibuster
See this sample [photos-of-the-year.com].
Then use the graph function to make a chart based on the earnings data.
The dip towards June was a smart pricing hit. That is the point I started blocking MFA's. Data from Jan 04 to Oct 05
[tinypic.com...]
Date range is mid December 2004 to date. Trend is 21 day average, dots are individual days.
Let say you have the highest earning $25 in Sep 2005, then the factor is 4 (=100/25), just multiple all earning value by 4 will scale it up.
This is becoming pretty difficult to compare if everyone is plotting a different date range...
Yes. Without scale, the graphs don't really tell us much. The squiggles could represent major changes or just random fluctuations.
-- Roger
Sept 04 to Current
I wish I could put another line in there to represent number of clicks.... wouldn't look so pretty.
Would it be against the TOS to put a line in there for clicks so long as no numbers are revealed?
Should we get some structure to this comparison. Set a certain date range and then compare?Would it be against the TOS to put a line in there for clicks so long as no numbers are revealed?
Dunno about TOS, but a line doesn't tell you very much.
I think CTR vs PI might be useful. If its not a line with
y=ax and a=0 APPROXIMATELY and you can exclude timeseries influences [aka you fiddling with the ads changing colours etc] you might be able to draw conclusions.
CTR vs eCPM should be fairly linear in a regression [use the right fitting curve though ;)]
ecpm vs time is interesting. If the regression is showing less wiggels at the end, the G algorithm is working, imo. If it bounces about the algo can't find an optimal fit for your set etc. There is soooo much you can do with that data ...
[tinypic.com...]
You think the pictures are "interesting"? Last night must have been a good night out!
Re trends, some sites seem to have experienced some significant growth in the last couple of months, but other sites experienced the same last year and then earnings fell back significantly this year. But I agree it is difficult to tell what is going on just from the graphs. It's like driving in the pouring rain when your windscreen wipers have broken: you know something is there but you probably won't find out what until you've hit it.
You think the pictures are "interesting"? Last night must have been a good night out!
I'm trying to be polite and show how intelligent I am:( Please don't blow my cover! Really, they all look like squiggles and dots!
I think general discussion of our analysis of trends we have noticed individually is more helpful - interpreting other's graphs with most of the detail missing is difficult. Probably about the only instance where a picture *doesn't* speak a thousand words.
Daily earnings as a 100% ratio thing for the last few months or so, dates given on the axis. I tweak the ads every week, so itd be difficult to draw any conclusions from it. Ive also just switched a lot to chitika this past couple days, which resets any trend analysis.
Does anyone have any tips on getting stuff into excel for analysis? At the moment I use a data query, download the csv to a fixed location, and just hit refresh in another xls file to pull in the values. Trouble is, it also pulls in the 'totals' row, and I need to go about updating all the source values for each of my graphs.
I like the idea of setting up a site to collect this 'anonymised' data, and am thinking of doing it. Would google be ok with it? Would people be willing to upload their stats, without faking them? Could I publish the URL here?
I would probably provide a xls file which would anonymise and standardise the data first for upload.
Does anyone have any tips on getting stuff into excel for analysis? At the moment I use a data query, download the csv to a fixed location, and just hit refresh in another xls file to pull in the values. Trouble is, it also pulls in the 'totals' row, and I need to go about updating all the source values for each of my graphs.
I use OpenOffice just import and press tab and you are done. Then I replace the annoying "," with ".", clean out the percentage into decimals in the CTR field and do a cut and paste into S-PLUS.
A free alternative to S-Plus is R although the data GUI is a bit cr@p, but it's nearly like S.
My biggest probem with Excel [besides the horrid Lego graphics] is in my opinion that it's more like a discrete package for accountants with hard numbers and really not build for highly stochastic time series data with multiple unknown covariates :\.
I mean you have the main influencing factors that is Google and Yahoo traffic then you have a weekday influence and a seasonal influence, then server speed backbone location, etc etc. Things like running means etc are probably valid in Excel but anything further is most likely too work intensive for a marketing demonstration/accountancy package. Perfectly reaonably if you want to do your accounts but useless if you have soft data.
There are also several fitting algorithms where you just need to press a button in a real statistics package, while you probably sit there for ages fighting with Excel to make ugly graphics that tell you eff all.
Rant over ;)
