... it could be all about buying habbits and trends.

Amazon has been pretty good in tracking hardware/software sales but not all is what it seems. Although it's a good indicator, conflict in some data between NPD and Amazon could suggest that Playstation guys are more likely to buy from Amazon to cause an offset. What I mean by this is if X1 and PS4 hardware or software were pretty much the same position on the Amazon monthly bestsellers, Xbox hardware/sofware would be ahead in the NPD.

The theory

Amazon data is offset in PS4s favour.

Reason for theory

Sofware sales between NPD and Amazon showed that Amazon monthly ranking had the PS4 version of numerous amounts of games ahead  while NPD revealed the X1 version winning since the launch of both consoles. It never, not once, showed the opposite. Also, even though Amazon and NPD matched winners of hardware for every month, it doesn't do against the theory. But on the hardware front it's too weak to use until we have a PS4 winning hardware on Amazon with NPD showing that the X1 won.

Explanation of how data being offset can show conflicting results in favour of one platform

I do not deny that but I'm just saying it's slightly offset. If PS fans do indeed prefer to buy on Amazon, let's say for argument sake, 50% of PS fans buy from amazon and 45% of xbox fans but from amazon. A games sell 110k on X1 and the equivalent game sells 100k on on PS4. On NPD it will show the X1 ahead. On amazon the PS4 version would of sold 50k while the X1 version would of sold 49.5k and be lower ranked. So it can sometimes show PS4 lead on Amazon but be behind on NPD but because of the way it's offset, it will hardly happen the other way around.

Why is this happening?

Originally I thought it could be to do with average age of fanbase being different. Because the Xbox fanbase used to be a lot younger, chances are they still are. Are the shopping habbits different for younger gamers? Do they prefer brick and mortar shops? Do they prefer gaming specific online retailers? Are teens less likely to have credit cards for online shopping? God knows! Someone mentioned about MS consistently having stronger X1 related deals outside of Amazon. Not sure that is true but another reason why it could offset in data on Amazon

To conclude, if you see the X1 a few place behind on the bestselling Amazon charts, don't be shocked to see it outsell the PS4 in the NPD data. Same is true for software.

Amazon data (change date in link to see other months): http://www.amazon.com/gp/bestsellers/2015-01/videogames/ref=zg_bsar_videogames_pg_3?ie=UTF8&pg=3#1

NPD (OP has link to all NPD threads): http://gamrconnect.vgchartz.com/thread.php?id=198105&page=1

I don't think anyone ever pretended it was entirely accurate. It's good for trends and trends usually determine which platform is preferred at the time. Which is the reason why it has always been correct this gen.

So why not use it to determine the winner of each NPD?

Your going to need more than 1 month of data to reach a conclusion like that. Plus a software purchase does not directly correlate with a sold console. So you need entirely different data set, solely consoles and bundles to say something about hardware.

Amazon is not 100% source, but sure more accurate than other site.

Hmm... But it hasn't been wrong yet.

Agree with  your opinion, but we all know this,  Amazon pre order data will never become a primary source of sales data,  because their number is not constant, it's pre order data not actual sales data. So it's bound to change, unless you are discussing sales data from Amazon not pre order data.  Pre order data only good for  projecting future sales.

Well it should always be taken with a grain of salt but it certainly has done a good job so far this gen

Confirmation Bias.

As of now, the only thing you can conclude is that a higher position on Amazon is likely to correlate with a higher NPD position since NPD is a retailer survey and Amazon is a Retailer.

I mean how exactly do you make the calculation in the following claim: "These 9 positions between the 2 in the top 20 should translate more than 40k between the 2 for the month." Unless you can reliably predict Future NPD results given Amazon ranks within a margin of error, then you haven't proven anything.

Not to mention you cannot make any reasonable conclusion about buying trends or habits with simply one month of data, use that conclusion to frame what occured in software, and then use that data set to somehow explain hardware, despite the fact that we can already assume that the correlation between hardware and software is not very strong considering the ratio is not one to one.

I wrote it from a subjective pov and said it from my own personal experience and used words like "should" intead of "will" etc. So no need to get on your high horse. I made it clear also.

And even though I used one month of data, I've noticed it in other months. But to start dragging in data from other months will take forever.

Take the theory for what it is. And yes I do think that if the data was a true representation 9 and 18 should equal more than 40k when upscaled to total sales from all stores. that 10k a week difference and when we assume that Amazon, having a guess, has 25% of gaming releated sales, we are talking about 2-3k difference from amazon. 2-3k difference a week from the largest store in the largest market should be a lot closer in the best selling rankings when you consider how many units the items in the top 20 are shifting.

It's been right for hardware. Software wise, no. NPD is a very good indicator but I feel that Amazon data needs to be offset to get a prediction for reasons explained in OP