[Putas]

Originally Posted by Mendel

To me it is an issue because it further obfuscates the reader from realizing how much the graph is actually skewed. In fact, I had a rescaled graph made where I added just 9 new lines to the left of the original graph, marked them 0.8x, 0.7x etc and was ready to call it a day when I took another close look, thought about it for a while and yelled WAIT A MINUTE! It's still skewed!

So at the first glance its majorly skewed, at a second glance I still didn't have a clear idea of just how skewed the graph actually was. I think I had to spend along the lines of 15 minutes of mspaint copy pasting and then thinking before I actually realized what the graph is supposed to look like in full scale.

Ahhh, I don't want to be personal but... you don't need to waste your time like that.
Some people learn one way and want to stay like it forever. Full scale is good to show absolute values. Point of this one was to show how one card is performing compared to other. Think like in full scale you have 0 for beginning, in this one performance of 2900 XT is used as zero. Just try. Suddenly everything will be simple and nice.

[Pete]

The point we're trying to make is that absolute values are a pretty good basis for clearly presenting relative performance--the only basis worth using if that's your intent, really. Unless you think something like this is more useful both in terms of most transparently conveying the data or best doing so in an at-a-glance format. The correct answer is no, a simple table would have been more accurate and concise. (OTOH, I think an agglomeration of data like this would be better served with a bar chart with the relevant %ages overlaid on the end of each bar.) To each his own, according to his intentions and limitations....

As for the implications of 2900 being zero, that makes comparison moot, no? Much simpler to make it one, which brings us back to my first paragraph.

[Mendel]

Originally Posted by Putas

Ahhh, I don't want to be personal but... you don't need to waste your time like that.
Some people learn one way and want to stay like it forever. Full scale is good to show absolute values. Point of this one was to show how one card is performing compared to other. Think like in full scale you have 0 for beginning, in this one performance of 2900 XT is used as zero. Just try. Suddenly everything will be simple and nice.

Think of it this way, skew the graph even further and at some point the graph doesn't really show anything meaningful, like my edited second picture of the nvidia graph. Is there any point, at which you would agree the graph is too skewed?

If you had 2900XT as zero or very close to it, you would probably have the bar like single pixel wide or not have any bar for it at all (what situation is this? some game the 2900xt can't run at all?), or , then if you only have bar for one product and nothing to compare it to, it doesn't make sense to have any graph at all and only the numerical data becomes of interest.

In one game in the graph they use 18 fps as starting point for the graph. Why? Why is 18 fps significant there? Why should anyone think 24 fps should feel like three times better than 20 fps? What is this difference indicative of?

In this case, to me, once I realized how skewed the graph was, the bars lost all meaning to me, the bars don't accurately describe the difference between two products unless rescaled. That said, I do give them some points back for including the actual framerate numbers. With them, there is some information in the graph left with which to work on.

[Putas]

I used zero to explain difference between this type of chart and full scale. As you can see here, people got so used to zero as referential point they refuse other ways. Yes, to have it allign to 1 is great thing, often better then then full scale. Which brings me again to disagreement. You can tell at glance percentage difference and have this relative difference for more games in one graph. It gets better with more games tested, you get easily average difference between the two. This one number, although not precise, is the one most wanted. Not card 1 have x fps and card 2 have y.
Your link is error 403.

[Mendel]

My link? What link?

edit: O I C you mean Pete's links

since you didn't answer my questions, I'd like to repose them

Putas:

1) Think of it this way, skew the graph even further and at some point the graph doesn't really show anything meaningful, like my edited second picture of the nvidia graph. Is there any point, at which you would agree the graph is too skewed?

2) In one game in the graph they use 18 fps as starting point for the graph. Why? Why is 18 fps significant there? What is this difference indicative of?

[Mendel]

1 picture tells 1000 words.

Putas, by your arguments, this picture should be about as ok as the first graph (though 2 game results less there)




However, here the picture is slightly more skewed, now there is absolutely no information in the graph except the numbers, the bars are pointless, agree?

well, to me, the bars are as pointless in the original graph, they tell nothing about the relative performance difference between the two cards.

(edited image quickly to add data below graph)

[Putas]

No, it is actually only about bars, fps numbers are irrelevant

[Pete]

Originally Posted by Putas

Your link is error 403.

That usually happens when one links directly to an inline image. Just click in the address bar and hit enter (or click the green 'go' arrow in Firefox).

[Putas]

Originally Posted by Mendel

However, here the picture is slightly more skewed, now there is absolutely no information in the graph except the numbers, the bars are pointless, agree?
(edited image quickly to add data below graph)

To be more specific- the bars carry still the same point, showing 3870 doing about 10% better then 2900 xt.

[AlphaWolf]

Originally Posted by Putas

To be more specific- the bars carry still the same point, showing 3870 doing about 10% better then 2900 xt.

you're not getting that information from looking at the bars...

[Putas]

True. You get that from bars+axis.

[Mendel]

Originally Posted by Putas

True. You get that from bars+axis.

If, and only if you disregard the axis, the bars hold no information alone, agree?

Is there any particular reason you avoid directly answering my questions in general?

[Putas]

Agree.
I did not answer since you were working with wrong premise.
So it would be something silly like 1) no 2) why not - it is not - nothing

[Mendel]

Originally Posted by Putas

Agree.

Thanks! It is my opinion that in such graph, bars alone should give a graphical representation of differences amongst the products that are being compared.

1) no

So if you look at the third and fourth picture of my original post, you could further skew the bars into oblivion and at no point it would be too skewed, not ridiculous in the slightest?



IMO it is okay to highlight differences, if you do include the starting point of the graph and then imply discontinuity in the graph if you want to cut "irrelevant" parts of it off. ( for layout purposes for example.)
This stuff is teached in high school math class, it's very well thought out, there are standards in how you make a graph for comparing things if you want to be informative. I expect to be treated as a person who wants information. I don't expect to be expected to use photoshop to make the graphs make sense to me

[Putas]

Yes, it would be ridiculous and ugly. More so since you excluded negative results. But similar reasons led to skewing- having results only in range of 90-120% why not crop it? It is not a big deal to add axis description, you should check for it everytime anyway. You cannot get rid of it with everytime. Do you have to imply discontinuity? I say no, not necessary, this is no exact science or engineering with its standards described in fat books. Just don't forgot they won't teach you everything in school and think for yourself- you will do fine without photoshop.

[neliz]

Originally Posted by Putas

Yes, it would be ridiculous and ugly. More so since you excluded negative results. But similar reasons led to skewing- having results only in range of 90-120% why not crop it? It is not a big deal to add axis description, you should check for it everytime anyway. You cannot get rid of it with everytime. Do you have to imply discontinuity? I say no, not necessary, this is no exact science or engineering with its standards described in fat books. Just don't forgot they won't teach you everything in school and think for yourself- you will do fine without photoshop.

the thing is that with cropping (which I agree, should be applied by default on most statistics.. that's the way I learned it) comes the loss of scale. That scale is the reason everyone talks about skewed graphs, simply because the scale of the original graph is gone.
Even if the axis are clearly labeled graphics are easier (and universal) to imprint the (now un-proportionally) big difference between 20 and 22 f/s.

that's why these graphs should scale properly. i.e. with fixed values for x and y (yes. graphs will become big comparing 200 f/s to 20 f/s) or scaling the graph so that the largest value is 100% and all the other results are scaled proportionally.

[Putas]

Originally Posted by neliz

that's why these graphs should scale properly. i.e. with fixed values for x and y (yes. graphs will become big comparing 200 f/s to 20 f/s) or scaling the graph so that the largest value is 100% and all the other results are scaled proportionally.

I still don't see a reason why graphs showed here should be called improperly scaled.
Your proposition would make comparsion between products harder.

[Mendel]

It's just the effect these graphs have when you first take a glance at them. It's this "Hey, that looks like a great improvement, no wait a minute..." factor that kind of offends people.

I don't think there's any problem layout wise using full scale graphs if one does them properly.

If we had a poll about which is better ( or preferred, more informative): the first graph of the first post or the version edited by me (given it some tweaks to make it not look like mspaint job) ... I'd bet good amounts of money that the edited version would be voted winner hands down!

Why? Because the starting point of the original graph is indicative of nothing (as even Putas admitted) whereas the meaning of starting point of full scale graph is self evident.

[Mendel]

Originally Posted by Putas

I still don't see a reason why graphs showed here should be called improperly scaled.

Because they don't fit the expectations of the viewers.

[Putas]

Originally Posted by Mendel

Why? Because the starting point of the original graph is indicative of nothing (as even Putas admitted) whereas the meaning of starting point of full scale graph is self evident.

I admitted under condition of missing description.
Your personal expectation does not set standard for other viewers.

[AlphaWolf]

Originally Posted by Putas

I admitted under condition of missing description.
Your personal expectation does not set standard for other viewers.

You seem to be the only one with a dissenting opinion, it might be more accurate to say its your expectation that does not set a standard for other viewers.

[neliz]

Originally Posted by AlphaWolf

You seem to be the only one with a dissenting opinion, it might be more accurate to say its your expectation that does not set a standard for other viewers.

hear hear, someone should run a poll and display the results in a non-skewed scale...

c'mon guys.. Putas is just trolling around..

[Geo]

With a name like "Putas"? Say it ain't so.

[neliz]

Originally Posted by Geo

With a name like "Putas"? Say it ain't so.

geh geh geh...

[Putas]

don't blame me for spanish language please

Originally Posted by neliz

c'mon guys.. Putas is just trolling around..

I am sorry for spoiling your trolling on innocent graph.