Less Wrong is a community blog devoted to refining the art of human rationality. Please visit our About page for more information.

Comment author: gjm 19 September 2017 10:22:23PM *  0 points [-]

serif readability is superior to that of sans-serif

Do you have actual solid evidence for that? I'm guessing that if you did you'd have given it already in your earlier comments, and you haven't; but who knows? (One of the answers to that Quora question mentions a study that found a small advantage for serifs. It also remarks that the difference was not statistically significant, and calls into question the choice of typefaces used, and says it's not a very solid study. So I hope you have something better than that.)

On a ~96ppi display [...] the apparent [...] font size that you need to bring out the superiority of serif typefaces is no larger than the minimum size called for by other accessibility guidelines.

Again, I would be interested in more information about what evidence you have about the font size required "to bring out the superiority of serif typefaces". For the avoidance of doubt, that isn't a coded way of saying "I bet you're wrong"; I would just like to know what's known about this and how solidly. I do not have the impression that these issues are as settled as you are making them sound; but I may just be unaware of the relevant work.

What OS is this on?

One instance is Firefox on Windows; the other is Firefox on FreeBSD (which I expect is largely indistinguishable in this context from Firefox on Linux). I concur with your guess that the people responsible for LesserWrong have not done thorough testing of their site on a wide variety of platforms, though I would be surprised if no one involved uses either Windows or Linux.

Yes, it looks tiny and hard to read.

LesserWrong has what looks to me like a weird multiplicity of different text sizes. Some of the text is clearly too small (personally I like small text, but I am aware that my taste is not universally shared). However -- and I must stress again that here I am merely describing my own experience of the site -- if I go to, say, this post on the Unix box at my desk right now then (1) the size of the type at my typical viewing distance is about the same as that of a decently typeset paperback book at its typical viewing distance, and (2) I find the text ugly and harder to read than it should be because various features of the typeface (not only the serifs) are poorly represented -- for me, on that monitor, after rendering by my particular machine -- at the available resolution. (The text is very similar in size and appearance to that on readthesequences.com; LW2.0 appears to be using -- for me, etc., etc. -- ETBembo Roman LF at 19.2px actual size, whereas RTS is using GaramondPrmrPro at 21px actual size. ETBembo has a bigger x-height relative to its nominal size and most lowercase letters are almost exactly the same size in each.)

Other issues, like color

Yup, agreed. But I would say the same about readthesequences.com even though its body text is black.

If every LW user looks at the site and says that, then we can't still conclude anything about serifs from that,

I agree. (Though it would, despite their hypothetical ignorance, be evidence. Someone who says "this text is hard to read because of the serifs" may be wrong, but I claim they are more likely to say it in the face of text that's hard to read because of its serifs than of text that's hard to read for some other reason.)

Perhaps I left too much implicit in my earlier comment, so let me try to remedy that. I firmly agree that the mere fact that some LW users believe some proposition about serifs in webpage text is perfectly compatible with the falsehood of that proposition. Even if it's quite a lot of LW users. Even if they have a lot of karma.

But the thing that actually matters here is not the general proposition about serifs, but a more specific question about the type used on LesserWrong. I wasn't equivocating between this and the general claim about serifs, nor was I unaware of the difference; I was deliberately attempting to redirect discussion to the more relevant point.

(Not that the general question isn't interesting; it is.)

[EDITED to add:] Of course much of what I wrote before was about the general proposition. Whether I agree with you about that depends on exactly what version of the general proposition we're discussing -- I take it you would agree with me that many are possible, and some might be true while others are false. In particular, I am somewhat willing to defend the claim that there are otherwise reasonable choices of text size for which typical seriffed typefaces make for a worse reading experience than typical sans-serif typefaces for people using 100ish-ppi displays, and that while this can be mitigated somewhat by very careful choice of serif typefaces and careful working around the quirks of the different text rendering users on different platforms will experience, selecting sans-serif typefaces instead may well be the better option. I am also willing to be convinced to stop defending that claim, if there is really good evidence against it.

Comment author: SaidAchmiz 19 September 2017 11:28:26PM 0 points [-]

Do you have actual solid evidence for that?

Not close at hand. You may reasonably consider my claim to be undefended for now. When I have the time, I'll try to put together a bit of a lit survey on this topic.

LesserWrong has what looks to me like a weird multiplicity of different text sizes. Some of the text is clearly too small (personally I like small text, but I am aware that my taste is not universally shared). However -- and I must stress again that here I am merely describing my own experience of the site -- if I go to, say, this post on the Unix box at my desk right now then (1) the size of the type at my typical viewing distance is about the same as that of a decently typeset paperback book at its typical viewing distance, and (2) I find the text ugly and harder to read than it should be because various features of the typeface (not only the serifs) are poorly represented -- for me, on that monitor, after rendering by my particular machine -- at the available resolution. (The text is very similar in size and appearance to that on readthesequences.com; LW2.0 appears to be using -- for me, etc., etc. -- ETBembo Roman LF at 19.2px actual size, whereas RTS is using GaramondPrmrPro at 21px actual size. ETBembo has a bigger x-height relative to its nominal size and most lowercase letters are almost exactly the same size in each.)

Right you are. The 16px size is what I saw on the front page.

Even on my machines, ET Book (source) does not seem to render as well as Garamond Premier Pro (in a browser).

Though it would, despite their hypothetical ignorance, be evidence. Someone who says "this text is hard to read because of the serifs" may be wrong, but I claim they are more likely to say it in the face of text that's hard to read because of its serifs than of text that's hard to read for some other reason.

I think this is literally true but relevantly false; specifically, I think this is false once you condition on the cause of the text's unreadability not being some gross and obvious circumstance (like, it's neon purple on a fuchsia background, or it's set at 2px size, etc.)

I think that someone who is ignorant of typography is no more likely to blame serifs in the case of the serifs being to blame than in the case of the text rendering or line length being to blame.

But the thing that actually matters here is not the general proposition about serifs, but a more specific question about the type used on LesserWrong. I wasn't equivocating between this and the general claim about serifs, nor was I unaware of the difference; I was deliberately attempting to redirect discussion to the more relevant point.

Noted. I was responding to the general claim.

As to the specific question, the matter of serifs is moot, because (as with all specific design decisions), each designer decision should be comprehensively user-tested and environment-tested, and as much user choice should be offered as possible.

Of course much of what I wrote before was about the general proposition. Whether I agree with you about that depends on exactly what version of the general proposition we're discussing -- I take it you would agree with me that many are possible, and some might be true while others are false.

Indeed.

In particular, I am somewhat willing to defend the claim that there are otherwise reasonable choices of text size for which typical seriffed typefaces make for a worse reading experience than typical sans-serif typefaces for people using 100ish-ppi displays … I am also willing to be convinced to stop defending that claim, if there is really good evidence against it.

Nope, the claim is reasonable. Websites where information density is more important than long-form readability, or where text comes in small chunks and a user is expected not to read straight through but to extract those chunks, may be like this. For that use case, a smaller point size of "body" text may be called for, and a well-chosen sans font may be a better fit.

LessWrong is not such a website, though a hypothetical LessWrong community wiki may be (or it may not be; it depends on what sort of content it mostly contains).

(Aside: I somewhat object to speaking of "typical" serif typefaces, because that's hard to resolve nowadays. I suspect that you know that, and I know that, but in a public discussion it pays to be careful with language like this.)

However:

very careful choice of […] typefaces and careful working around the quirks of the different text rendering users on different platforms will experience

… is always advisable, regardless of typographic or other design choices.

Comment author: Viliam 19 September 2017 10:39:33PM 0 points [-]

When I compare the two examples, the second one feels "clear", while the first one feels "smudgy". I have to focus more to read the first one.

Comment author: SaidAchmiz 19 September 2017 11:00:54PM *  2 points [-]

1. OS (and version), browser (and version), device/display, etc.?

(General note: folks, please, please include this information whenever you say anything to a web designer/developer/etc. about how a website looks or works for you!!)

2. Great! If one of them feels clear, then this goes to show exactly what I was saying: user choice is good.

Comment author: Jiro 19 September 2017 08:34:04PM 0 points [-]

People are clearly posting things there that postdate the DB import, so they must be logging in. Also, that doesn't explain it working better on Chrome than on other browsers.

Comment author: SaidAchmiz 19 September 2017 08:59:28PM 1 point [-]

A private beta has been ongoing, clearly.

Comment author: gjm 19 September 2017 11:53:04AM 6 points [-]

Here's what I think is the conventional wisdom about serif/sans-serif; I don't think it is in any way contradicted by the material you've linked to.

Text that is small when measured in display pixels is generally harder to read fluently when set in a typeface with serifs.

Only interested in readers with lovely high-DPI screens? Go ahead, use serifs everywhere; it'll probably be fine. Writing a headline, or a splash screen with like 20 words on it? Use serifs if they create the effect you want; the text won't be small enough, nor will there be enough of it in a block, for there to be a problem.

But if you are choosing a typeface for substantial chunks of text that might be read on a not-so-great screen, you will likely get better results with a sans-serif typeface.

So, what about those domain experts? Jakob Nielsen is only addressing how things look on "decent computer screens with pixel densities of 220 PPI or more". Design Shack article 1 says that a blanket prohibition on serifed typefaces on screens is silly, which it is. But look at the two screenshots offered as counterexamples to "Only use serifs in print". One has a total of seven words in it. The other has a headline in a typeface with serifs ... followed by a paragraph of sanf-serif text. Design Shack article 2 says that sans-serif typefaces are better "for low-resolution displays", though it's not perfectly clear what they count as low-resolution. The Quora question has a bunch of answers saying different things, mostly not from "domain experts" in any strong sense.

I like seriffed typefaces. In a book, sans-serif is generally hideous and offputting to me. On my phone or my laptop, both of which have nice high-resolution displays, Lesser Wrong content with serifs looks just fine. (Better than if it were set sans-serif? Dunno.) On the desktop machine I'm using right now, though, it's ugly and it feels more effortful to read than the corresponding thing on, say, Less Wrong. For me, that is.

now we have two people agreeing [...] Does that make it more true?

Yes. More precisely: the proposition we should actually care about here is not some broad generality about serif versus sans-serif typefaces, but something like "Users of Lesser Wrong will, on the whole, find it a bit easier on the eyes if content is generally set in sans-serif typefaces". Consider the limiting case where every LW user looks at the site and says "ugh, don't like that font, the serifs make it harder for me to read". Even if all those users are shockingly ignorant of typography, this is a case where if no one likes it, then it is ipso facto bad.

Of course we don't have (anything like) the entire LW community saying in chorus how much they dislike those serifs. But yes, when what matters is the experience of a particular group of people, each individual person who finds a thing bad does contribute to its badness, and each individual person who says it's bad does provide evidence for its badness.

What amount of karma and local agreement does it take to get to the truth?

Karma is relevant here only as a proxy for participation. A crude answer to this question is: enough to constitute a majority of users, weighted by frequency of use.

In case I haven't made it clear enough yet, I am not arguing that LW people are always right, or that high-karma LW people are always right. I am arguing that when the thing at issue is the experience of LW people, the experiences of LW people should not be dismissed. And I am arguing that on the more general question (are typefaces with serifs a bad idea on the web?) the simple answer "no; that's an outdated bit of bogus conventional wisdom" is in fact just as wrong as the simple answer "yes; everyone knows that".

Comment author: SaidAchmiz 19 September 2017 05:32:13PM *  1 point [-]

And I am arguing that on the more general question (are typefaces with serifs a bad idea on the web?) the simple answer "no; that's an outdated bit of bogus conventional wisdom" is in fact just as wrong as the simple answer "yes; everyone knows that".

Disagree. (Keep reading for details.)

But if you are choosing a typeface for substantial chunks of text that might be read on a not-so-great screen, you will likely get better results with a sans-serif typeface.

This is still incorrect, because serif readability is superior to that of sans-serif, and see below for the matter of "not-so-great screens".

Screen DPS

Given the pixel resolution per character you need to make serifs work, they are inferior on the screen… if you have a 72ppi (or less) display.

Now, such displays exist; here's one. They are quite rare, though, and designed for entertainment, not work. The idea that any appreciable percentage of LW users have such hardware seems implausible.

On a ~96ppi display (such as this nearly decade-old cheap flat-panel I'm using right now, or indeed any display display made in the past 15+ years), the apparent (angular, a.k.a. "CSS reference pixel") font size that you need to bring out the superiority of serif typefaces is no larger than the minimum size called for by other accessibility guidelines.

“The LW 2.0 font is less readable”

On the desktop machine I'm using right now, though, it's ugly and it feels more effortful to read than the corresponding thing on, say, Less Wrong. For me, that is.

1. What OS is this on? If the answer is "Linux" or "Windows", then part of the answer is "text rendering works very different on those operating systems, and you have a) test your site on those systems, b) make sure to make typographic choices that compensate, c) take specific actions to ensure that the user experience is adjusted for each client platform". I of course can't speak to (a), but (b) and (c) are not in evidence here.

2. The body text font size on LW 2.0 is too small (especially for that font), period. Again I refer you to https://www.readthesequences.com/Biases-An-Introduction; the body text is at 21px there. I consider that to be a minimum (adjusted for the particular font); whereas LW 2.0 (with a similar type of font) is at 16px. Yes, it looks tiny and hard to read. (But have you tried zooming in? What happens then?)

3. Other issues, like color (#444, in this case) affecting text rendering. I speak of this in my other comments.

“Consensus matters locally”

Consider the limiting case where every LW user looks at the site and says "ugh, don't like that font, the serifs make it harder for me to read". Even if all those users are shockingly ignorant of typography, this is a case where if no one likes it, then it is ipso facto bad.

If every LW user looks at the site and says that, then we can't still conclude anything about serifs from that, because if all of those users have not the smallest ounce of typography or design expertise, then they don't know what the heck they like or dislike, serif-wise.

Let me be clear: I'm not saying that people can't tell whether they like or dislike a particular thing. I am saying that without domain knowledge, people can't generalize their preferences. Ok, so some text on their screen is hard for them to read. What's making it so? The fact that a font has serifs? Or maybe just that it's a particular kind of serif font? Or the font weight? Or the weight grade? Or the shape of the letterforms (how open the curves are, for instance, or the weight variability, perhaps due to which "optical size" is being used)? Or the color? Or the subpixel rendering settings? Or the kerning? Or the line spacing? Or the line length? Or the text-rendering CSS property setting? If you (the hypothetical-user you) don't know what most or all of those things are, then sure your preferences are real, but your opinion (generalized from those preferences) is worth jack squat.

In other words: "if no one likes it, then it is ipso facto bad"—yes, but what, exactly, is "it"? You're equivocating between two meanings, in that sentence! So, this is true:

“If no one likes <a particular specific thing>, then <that specific particular thing> is bad.”

Yes. Granted. But you seem to want to say something like:

“If no one likes <a particular specific thing>, then <things in a class that include that specific particular thing> are bad.”

But any particular thing belongs to many different classes, which intersect at the point defined by that thing! Obviously not all those classes are ipso facto bad, so which one(s) are we talking about?? We have no idea!

I am arguing that when the thing at issue is the experience of LW people, the experiences of LW people should not be dismissed.

Dismissed? No. Taken at anything even remotely resembling face value? Also no.

Come on, folks. This is just a rehash of the "people don't have direct access to their mental experience" debate. You know all of this already. Why suddenly forget it when it comes up in a new domain?

Comment author: Habryka 16 September 2017 11:20:43PM 1 point [-]

I am slightly hesitant to force authors to think about how their posts will look like in different fonts, and different styles. While I don't expect this to be a problem most of the time, there are posts that I write where the font choice would matter for how the content comes across.

Medium allows the writer to chose between a sans-serif and a serif font, which I like a bit more, but I would expect would not really satisfy Alicorn's preferences.

Maintaining multiple themes also adds a lot of design constraints and complexity to updating various parts of the page. The width of a button might change with different fonts, and depending on the implementation, you might end up needing to add special cases for each theme choice, which I would really prefer to avoid.

Overall, my hypothesis is that Alicorn might not dislike serif-fonts in general, but might be unhappy about our specific choice of serif fonts, which is indeed very serify. I would be curios whether she also has a similar reaction to the default Medium font, for example displayed in this post: https://medium.com/@pshrmn/a-simple-react-router-v4-tutorial-7f23ff27adf

Comment author: SaidAchmiz 18 September 2017 08:19:39AM *  2 points [-]

Update: I have a recommendation for you!

Take a look at this page: https://wiki.obormot.net/Reference/MerriweatherFontsDemo

The Merriweather and Merriweather Sans fonts (available for free via Google Fonts) are, as you can see, designed to be identical in width, line spacing, etc. They are quite interchangeable, in body text, UI, etc. Both are quite readable, and aesthetically pleasing.

(As a bonus, active on that page is a tiny bit of JavaScript trickery that sets different body text font weights depending on whether the client is running on a Mac, Windows, or Linux platform, to ensure that everyone sees basically the same thing, and enjoys equally good text readability, despite differences in text rendering engines. Take a look at the page source to see how it's done!)

UPDATE 2: A couple of mockups (linking to individual images because Imgur's zoom sucks otherwise). Be sure to zoom in on each image (i.e. view at full magnification):

LW 2.0 with Merriweather:

LW 2.0 with Merriweather Sans:

Comment author: 9eB1 17 September 2017 02:47:31PM 0 points [-]

Sure.

Since then I've thought of a couple more sites that are neither hierarchical nor tag-based. Facebook and eHow style sites.

There is another pattern that is neither hierarchical, tag-based nor search-based, which is the "invitation-only" pattern of a site like pastebin. You can only find content by referral.

Comment author: SaidAchmiz 17 September 2017 05:56:55PM 1 point [-]

It is therefore not a coincidence that Facebook is utterly terrible as a content repository. (I am unfamiliar with eHow.)

Comment author: 9eB1 17 September 2017 03:13:06AM 0 points [-]

That is very interesting. An exception might be "Google search pages." Not only is there no hierarchical structure, there is also no explicit tag structure and the main user engagement model is search-only. Internet Archive is similar but with their own stored content.

With respect to TV Tropes, I'd note that while it is nominally organized according to those indexes, the typical usage pattern is as a sort of pure garden path in my experience.

Comment author: SaidAchmiz 17 September 2017 04:11:51AM 0 points [-]

With respect to TV Tropes, I'd note that while it is nominally organized according to those indexes, the typical usage pattern is as a sort of pure garden path in my experience.

I have encountered a truly shocking degree of variation in how people use TVTropes, to the extent that I've witnessed several people talking to each other about this were each in utter disbelief (to the point of anger) that the other person's usage pattern is a real thing.

Generalizations about TVTropes usage patterns are extremely fraught.

Comment author: Habryka 17 September 2017 02:51:10AM 1 point [-]

Thanks for the recommendations!

"This is a slightly odd comment, if only because "hierarchical or tag-based structures" describes almost all extant websites that aggregate / archive / collect content in any way!"

Well, the emphasis here was on the "more". I.e. there are more feed based architectures, and there are more taxonomy/tagging based architectures. There is a spectrum, and reddit very much leans towards the feed direction, which is what LessWrong has historically been. And wiki's very much lean towards the taxonomy spectrum. I feel we want to be somewhere in between, but I don't know where yet.

Comment author: SaidAchmiz 17 September 2017 04:08:01AM *  3 points [-]

Certainly there is variation, but I actually don't think that viewing that variation as a unidimensional spectrum is correct. Consider:

I have a blog. It functions just like a regular (wordpress) blog—it's sequential, it even has the usual RSS feed, etc. But it runs on pmwiki. So every page is a wikipage (and thus pages are organized into groups; they have tags and are viewable by group, by tag, by custom pagelist, etc.)

So what is that? Feed-based, or tag-based, or hierarchical, or... what? I think these things are much more orthogonal than you give them credit for. Tag-based structure can overlay hierarchical structure without affecting it; custom pagelist/index structure, ditto; and you can serve anything you like as a feed by simply applying an ordering (by timestamp is the obvious and common one, but there are many other possibilities), and you can have multiple feeds, custom feeds, dynamic feeds, etc.; you can subset (filter) in various ways…

(Graph-theoretic interpretations of this are probably obvious, but if anyone wants me to comment on that aspect of it, I will)

P.S.: I think reddit is a terrible model, quite honestly. The evolution of reddit, into what it is today, makes it fairly obvious (to me, anyway) that it's not to be emulated.

Edit: To be clear, the scenario above isn't hypothetical—that is how my actual blog works.

Edit2: Consider also https://readthesequences.com. (It, too, runs on pmwiki.) There's a linear structure (it's a book; the linear navigation UI takes you through the content in order), but it would obviously be trivial to apply tags to pages, and the book/sequence structure is hierarchical already.

Comment author: Alicorn 17 September 2017 01:19:57AM 3 points [-]

I do not like the readthesequences font. It feels like I'm back in grad school and also reading is suddenly harder.

I'm on a Mac 'fox.

Comment author: SaidAchmiz 17 September 2017 02:37:50AM *  3 points [-]

Ok, thanks!

FYI, your assessment is in the extreme minority; most people who have seen that site have responded very positively to the font choice (and the typography in general). This suggests that your preferences are unusual, in this sphere.

I say this, not to suggest that your preference / reaction is somehow "wrong" (that would be silly!), but a) to point out the the danger in generalizing from one's own example (typical mind blah blah), and b) to underscore the importance of user choice and customization options!

rest of this response is not specifically for Alicon but is re: this whole comment thread

This is still a gold standard of UX design: sane defaults plus good[1] customizability.

[1] "Good" here means:

  • comprehensive
  • intuitive
  • non-overwhelming (i.e. layered)

Note, these are ideals, not basic requirements; every step we take toward the ideal is a good step. So by no means should you (the designer/developer) ever feel like "comprehensive customizability is an unreachable goal; there's no reason to bother, since Doing It Right™ is too much effort"! So in this case, just offering a couple of themes, which are basic variations on each other (different-but-matching font choices, a different color scheme), is already a great thing and will greatly improve the user experience.

Comment author: DragonGod 17 September 2017 01:03:26AM 2 points [-]

simply don't suffice to conclude that LessWrongers are likely to have a finger on the pulse of arbitrary domains of knowledge/expertise

Do you suggest that the consensus on Lesswrong about arbitrary domains is likely to be true with P <= 0.5?
As long as Pr(B|lesswrong consensus) is > 0.5, then Lesswrong consensus remains Bayesian evidence for truth.

Comment author: SaidAchmiz 17 September 2017 01:14:31AM 1 point [-]

Do you suggest that the consensus on Lesswrong about arbitrary domains is likely to be true with P <= 0.5?

For some domains, sure. For others, not.

We have no real reason to expect any particular likelihood ratio here, so should probably default to P = 0.5.

View more: Next