4 ms·

It's the same trick that Apple and Dell use. The left side of the graph is not 0% but out of habit we think it is.

by fxtentacle 3y ago

pacifika 3y ago
What’s the left side offset? It makes no sense to me at any number
- broast 3y ago
  The post suggests it is 40% which seems correct to me.
  - loeg 3y ago
    From other comments, I believe it's 33%: https://news.ycombinator.com/item?id=36600945 https://news.ycombinator.com/item?id=36600945
jerf 3y ago
"The left side of the graph is not 0% but out of habit we think it is." This is a learned behavior and it can be fixed. I'm mostly broken of it; I assume it's not until I see that it is. Only mostly, not claiming to be a bias-free superhuman here, but as a for instance I had to check the comments to see what was offending people about this chart because I didn't assume it was anchored at zero in the first place. I recommend making the effort if you can; most charts out of school aren't anchored at zero, many wouldn't even be useful if they were. What jumped out at me actually was the lack of units. At least it links to the original report, and I think this is the Lighthouse Performance Score, which... is itself in units I don't understand? It's some sort of percentage, not speed measure, I think... looks like percentage conformance to an SEO/performance test suite? I'm not sure this is really something that can be bar graphed in the first place sensibly. Percentage conformance to test suites is at least a marginal case at best for a bar graph... how "numeric" such a quantity is is very debatable. e.g., a 40% compliant product may in fact be "less" compliant than a 30% compliant product if the former is compliant on a whole bunch of incidental check points but misses the most important ones, where the latter is compliant on the important ones and misses the incidentals, so such "percentages" can break simple numeric properties like comparability or the transitivity of such comparisons and really aren't numbers. Compliance to such a test suite will also be highly dependent on how you use the system; odds are its competitors can be tuned to go higher without much work and that very easy-to-make mistakes will send a Astro-based site plummeting because of what a user might naturally do with it, through no fault of the framework.
- yjftsjthsd-h 3y ago
  > most charts out of school aren't anchored at zero, many wouldn't even be useful if they were. What would make it useless to start a graph at zero?
  - jerf 3y ago
    Consider a chart of the recent M2 money supply for the last couple of years: https://fred.stlouisfed.org/graph/?g=16MIK https://fred.stlouisfed.org/graph/?g=16MIK Whether that is "useless" depends on your point. If you want to make a point that they are not large percentage fluctuations you may prefer to anchor it at zero. But if you want to talk about the fluctuations themselves, you want the chart to be full of those fluctuations, not anchored at zero and all the fluctuations just little wiggles relegated to the top. So it does depend on what you want to do, but there are use cases where anchoring at zero is either useless, or even itself at times deceptive, especially in cases where "zero" is not a particularly distinguished value on its own. (e.g., voltage. zero voltage is itself a relative measure to some standard of voltage, so a voltage chart being anchored to zero may not mean anything if zero voltage in some context isn't special but is itself somewhat arbitrary.)
    - zarzavat 3y ago
      If you want to make a chart about the fluctuations themselves, another way would be to show the differences on chart with a centerline at zero and bars that go positive and negative. Or if you really want the cumulative then normalize the first or last data point to be zero. The problem that people have with not starting at zero is how arbitrary it is. You can tell a completely different story depending on where you choose the starting point. So there should be some kind of justification for the chosen Y range.
      - jerf 3y ago
        I don't know 100% if you think you're disagreeing with me or not, but I'd consider your second paragraph a restatement of my point from a very different (and still valid) point of view. It's always something that a graph maker should be thinking about for sure, and a blanket "anchor it at zero" is neither desirable, nor even necessarily possible all the time.
dandellion 3y ago
I just assume that if the axes or the columns of a graph are not labelled or they don't have units the graph is worthless and whoever made it is either incompetent or worse, trying to manipulate people.
bombcar 3y ago
How to Lie with Charts - https://www.amazon.com/How-Charts-Gerald-Everett-Jones/dp/1419651439 https://www.amazon.com/How-Charts-Gerald-Everett-Jones/dp/14... or How to Lie with Stastitics - https://www.amazon.com/How-Lie-Statistics-Darrell-Huff/dp/0393310728 https://www.amazon.com/How-Lie-Statistics-Darrell-Huff/dp/03... Always good reading, find at a library near you!

It's the same trick that Apple and Dell use. The left side of the graph is not 0% but out of habit we think it is.

pacifika 3y ago

What’s the left side offset? It makes no sense to me at any number

broast 3y ago
The post suggests it is 40% which seems correct to me.
- loeg 3y ago
  From other comments, I believe it's 33%: https://news.ycombinator.com/item?id=36600945 https://news.ycombinator.com/item?id=36600945

jerf 3y ago

"The left side of the graph is not 0% but out of habit we think it is." This is a learned behavior and it can be fixed. I'm mostly broken of it; I assume it's not until I see that it is. Only mostly, not claiming to be a bias-free superhuman here, but as a for instance I had to check the comments to see what was offending people about this chart because I didn't assume it was anchored at zero in the first place. I recommend making the effort if you can; most charts out of school aren't anchored at zero, many wouldn't even be useful if they were. What jumped out at me actually was the lack of units. At least it links to the original report, and I think this is the Lighthouse Performance Score, which... is itself in units I don't understand? It's some sort of percentage, not speed measure, I think... looks like percentage conformance to an SEO/performance test suite? I'm not sure this is really something that can be bar graphed in the first place sensibly. Percentage conformance to test suites is at least a marginal case at best for a bar graph... how "numeric" such a quantity is is very debatable. e.g., a 40% compliant product may in fact be "less" compliant than a 30% compliant product if the former is compliant on a whole bunch of incidental check points but misses the most important ones, where the latter is compliant on the important ones and misses the incidentals, so such "percentages" can break simple numeric properties like comparability or the transitivity of such comparisons and really aren't numbers. Compliance to such a test suite will also be highly dependent on how you use the system; odds are its competitors can be tuned to go higher without much work and that very easy-to-make mistakes will send a Astro-based site plummeting because of what a user might naturally do with it, through no fault of the framework.

yjftsjthsd-h 3y ago
> most charts out of school aren't anchored at zero, many wouldn't even be useful if they were. What would make it useless to start a graph at zero?
- jerf 3y ago
  Consider a chart of the recent M2 money supply for the last couple of years: https://fred.stlouisfed.org/graph/?g=16MIK https://fred.stlouisfed.org/graph/?g=16MIK Whether that is "useless" depends on your point. If you want to make a point that they are not large percentage fluctuations you may prefer to anchor it at zero. But if you want to talk about the fluctuations themselves, you want the chart to be full of those fluctuations, not anchored at zero and all the fluctuations just little wiggles relegated to the top. So it does depend on what you want to do, but there are use cases where anchoring at zero is either useless, or even itself at times deceptive, especially in cases where "zero" is not a particularly distinguished value on its own. (e.g., voltage. zero voltage is itself a relative measure to some standard of voltage, so a voltage chart being anchored to zero may not mean anything if zero voltage in some context isn't special but is itself somewhat arbitrary.)
  - zarzavat 3y ago
    If you want to make a chart about the fluctuations themselves, another way would be to show the differences on chart with a centerline at zero and bars that go positive and negative. Or if you really want the cumulative then normalize the first or last data point to be zero. The problem that people have with not starting at zero is how arbitrary it is. You can tell a completely different story depending on where you choose the starting point. So there should be some kind of justification for the chosen Y range.
    - jerf 3y ago
      I don't know 100% if you think you're disagreeing with me or not, but I'd consider your second paragraph a restatement of my point from a very different (and still valid) point of view. It's always something that a graph maker should be thinking about for sure, and a blanket "anchor it at zero" is neither desirable, nor even necessarily possible all the time.

dandellion 3y ago

I just assume that if the axes or the columns of a graph are not labelled or they don't have units the graph is worthless and whoever made it is either incompetent or worse, trying to manipulate people.

bombcar 3y ago

How to Lie with Charts - https://www.amazon.com/How-Charts-Gerald-Everett-Jones/dp/1419651439 https://www.amazon.com/How-Charts-Gerald-Everett-Jones/dp/14... or How to Lie with Stastitics - https://www.amazon.com/How-Lie-Statistics-Darrell-Huff/dp/0393310728 https://www.amazon.com/How-Lie-Statistics-Darrell-Huff/dp/03... Always good reading, find at a library near you!