Hacker News Reader

Comment by trimbo | Hacker News Reader

> They recruited 63 healthy individuals, 135 patients with mild cognitive impairment such as memory loss, and 68 people with Alzheimer's – the most prevalent type of dementia – in four hospitals in Barcelona and the metropolitan area

Wait so, N = 266? All from Barcelona?

C'mon.

View on HN

With proper random sampling you only need something like N = 27 (I forget exactly) to reach p = 0.05

Assuming random sampling, it's a misconception that you need ten thousand participants to have a useful study.

Cue “p = 0.05 is a random choice” and a big thread about p-hacking. Any takers?

It feels like everything to say has already been said about the bias of small sample sizes and the dangers of p-hacking. But it also seems like the discussion is inevitable here.

I wonder if there’s a new insight to glean, somehow...

For a background on why p-values are misleading, see https://youtu.be/42QuXLucH3Q which is excellent. He points out that 0.05 was selected by Ronald Fischer in 1925, and people have gone with it since.

The reason we have stuck with it ever since it s because it's a decent cutoff. I think in physics they sometimes use a much higher threshold of 5 sigma. Presumably because having much more data means they can afford to.

A 5% false positive rate can be pretty high depending on the situation, like whether or not a drug is effective.

Isn't the crux just that, "proper random sampling" can be quite hard. Assumptions are fine in theory, in reality, they can seriously throw us for a loop.

Right, so you really need to review the study's methodology to determine if N = 266 is actually enough or not. I am guessing my original parent did not conduct such a review.

I'm not sure which side to take on this issue. I think both can be right, but I lean towards agreement with the root comment.

Root parent comment is right to question if the N=266 is truly random given the description we do have at hand.

You are right to say N=266 could be truly random and we don't have enough information to definitively decide either way.

We can't read the study so it's just what we have - saying you have a truly random sample with a group just from Barcelona is extremely hard to accept. I mean even you've got to admit that.

It seems that his objection is he "homogeneity" of the population, they're all from Barcelona, if the sample had different locations I think he wouldn't object the small number( e.g N = 30 ).

Is not just the sample. I'm worried also by the homogeneity of the researchers

Your mixing up your statistics. Perfect random sampling of a normal distribution means you can sample 27 times and your statistical analysis is close to that of the entire population.

It's highly dependent on the magnitude of the effect (when comparing two populations) and the distribution across a population.

For something dementia, where even measuring the level of cognitive decline is highly subjective, no way N=27 would be enough.

As long as measurement errors are not significantly correlated with the tested "cause"/category, it should not have significant bearing on the results or their validity. In such a case, 27 individuals should be sufficient.

That being said, article mentions 63+135+68 individuals in 4 hospitals in a single city.

Also, you only need about 400 samples for 95% confidence on a proportion at the highest variance point of p1 = . 0.50. It used to drive me nuts when my CEO used to say “10 clicks out of 10,000 on this A/B test is in the noise”. No, that’s not in the noise, it’s almost certainly a clickthrough rate of close to 0.

Can you explain how N=27 could possibly represent and be extended to billions of humans? It is certainly not intuitive to me.

Assuming you take 27 people from all over the world at random, and all 27 of them have two legs, you can be fairly certain that most humans have two legs. You might also conclude that about 50% are male and about 60% are from Asia, but as traits get less frequent your error bars get bigger and bigger.

Of course actually picking 27 random people is extremely hard, so actual studies compensate by picking more people.

Random does not mean that it's enough to do a proper group comparison, because individual differences, genetics, lifestyle, etc... are additional factors to take in account for dementia related outcomes.

> With proper random sampling you only need something like N = 27 (I forget exactly) to reach p = 0.05

will be difficult to demonstrate that they can control for all individual differences at n=266 to prove differences in dementia outcomes are linked to a single variable.

Yes, I haven’t read the paper but how can you control for correlated stuff here, e.g. bilingual people are likely to be richer, more intelligent, exercise more, eat better more varied food and have richer social lives. Maybe.

Are immigrants and their children likely to be richer, eat better, etc?

Those who are bilingual, yes.

> Those who are bilingual, yes.

Wouldn't immigrants by nature have to be bilingual because they have to be able to communicate in the nation they immigrated to?

Go out. Look around. Talk to some people about their parents.

At least in the US, there are millions of people who don't speak English well or at all. Source: https://www2.census.gov/library/publications/2014/acs/acs-26...

There's a lot of immigration from Latin America though.

People that speaks fluent clingon must appear in the Forbes list then.

Someday I'd like to hear the lines

    - What says Lord Stanley? will he bring his power?

    - My lord, he doth deny to come.

    - Off with his son George's head!

in the original Klingon. Pursuing the synergy-recognising creative destruction of a Stanley[1] is no doubt in the wheelhouse of some on on the Forbes list.

[1] https://en.wikipedia.org/wiki/Battle_of_Bosworth_Field#Stanl...

There would need to be research done, nothing has been said if these people were all immigrants to Barcelona or the wages they were on, my guess is they are probably not truly random and replication will be difficult if we chose say bilingual in Wales, for example.

This study was not about an immigrant population: "we looked at whether in Barcelona – where everyone is more or less bilingual..."

There is not reason to think that inmigrant population is over-represented in the hospitals. Would be more logical to expect the opposite, in fact, as some immigrants could be more reluctant to go to the hospitals.

If I were to immigrate to Barcelona I would learn (Galician) Spanish, but perhaps not Catalan. (Though it’s a beautiful language.) I guess I would count as monolingual in their study.

Galician Spanish? Do you mean Castillian Spanish (which is another name for normal spanish)? Galician is a separate language

Thank you, I meant Castillian Spanish.

You can speak Spanish tainted by Galician words and modisms also. Is very common in Spanish speakers. The whole communication can be a 90% spanish and 10% Galician (or the opposite).

This is why they chose Barcelona, where the use of 2 languages (Catalan and Spanish) was "highly variable" and everyone in the city spoke or at least understood 2 languages:

"We wanted to take advantage of this variability, and instead of comparing monolingual and bilingual people, we looked at whether in Barcelona – where everyone is more or less bilingual – there was a certain degree of bilingualism that had neuroprotective benefits."

I think this doesn’t disprove the theory I’ve set out that people who know more languages than the average local population are likely to have other markers correlated with lower Alzheimer’s risk.

It's not about knowing more languages (I would be surprised to learn that they only knew two, after all it's Europe). This paper instead studies differences between people who actually use multiple languages, against people who don't.

There's still corellation here, like professions where the lingua franca isn't the local language (programming, medicine, music (to some degree), sciences, etc). The paper claims the effect appears equally strong in all occupations though.

But that's not the case here, Barcelona is a bilingual city. The average population speaks both Catalan and Spanish.

Very obvious propaganda to promote an agenda, not more, not less.

And don't make me start to talking about how normal is the mental health of the same people that trashed and burned Barcelona last years, because they felt "oppressed" by a man that died 30 years before they born. Or the lobby trying to erase the history claiming than Leonardo Da Vinci, Cervantes and "put your name here" were really from Catalonia. Very sane behavior, for sure.

Trust me (or your gut if you prefer). You can find crazy people equally among bilingual, monolingual, translingual and people with just mischieving tongues. Just take a look at how mental asylums are distributed.

This is just one of thousands of articles that support a patriotic narrative so are published. Is politics disguised as science. The opposite idea would never reach the journals.

Between Catalan, Basque, and Galician, many people in Spain are bilingual (if not trilingual since English is quite common). Rich, poor, intelligent, idiot, etc.

Have you got a source for any of that? I imagine a lot of demographic factors could confound such a study, but I don’t see why multilingual people would be richer/healthier/more educated. In a lot of western countries, I’d imagine a huge amount of the multilingual population would be made of less well off immigrants.

in singapore, it is the opposite.

the working class tend to be multilingual - nurses, deliverymen, (food) hawkers, construction workers, odd-job labourers. when you think of someone who has to juggle multiple languages - mandarin, cantonese, malay, tamil, a smattering of english - often as a necessity, it's someone who's had to hustle at the lower rungs of society.

the educated class tend to be monolingual - researchers, policy makers, lawyers, bankers - they've only ever had to operate in environments where the only language used is english.

I’m pretty sure the only places where being monolingual is common, are the countries where English is the dominant language. If you look at the whole population of the world, it’s more common to know more than one language than it is to know only one. If you’re born in a place that speaks a language other than English, then learning English is very common, and a lot of places (like Singapore) have multiple common languages, where people often speak more than one. Singaporeans will tend to speak English, and the language of their ethnic heritage. Plenty of Malaysians speak Malaysian and Chinese or English. Most people in Indonesia speak Indonesian and their local regional language. A lot of people in India speak their regional language and English or Hindi...

I’m not sure it’s a sign of social status in English speaking countries. The wealthier you are, the more you can just expect everybody around you to speak English. Where I live, the most common language I’d expect somebody to know besides English is Spanish, and the local Spanish speaking population isn’t known for their excessive wealth.

No. In backwater countries like Poland English is caught at school, but people don't speak or write in it. Many still prefer polish-language media because LACK CONFIDENCE in their english. It typically goes like "Oh, I had English 4 years ago at school, I'm not that good". You use English skills to read a label on a shampoo or an instruction manual. English is not spoken or even written regularly. Before the pandemic I even played board games with lots of people younger than me, largely university students, and there's visible hesitation to use an untranslated game.

There is some evidence that Poland was spared from XIII black plague, and - so far - relatively unaffected - by the covid19 pandemic - because it lies off the main transport and travel routes.

In Poland, most movies have a Polish voice-over. If you see someone using subtitles, you can be pretty sure he's a cinema fan. Back in the transitional period after 1989 it was common to find english labels on imported goods, but not anymore. Everything except radio music is translated.

I’m not sure what you think you’re going to prove by point out one example (which you yourself describe as being a comparatively isolated country).

The total population of multilingual people is obviously quite hard to measure. But every study I’ve seen has put it above 50% of the worlds population, with English being the most common second language choice. EU stats on Poland say 1/4 of its population speak a second language well enough to hold a conversation, with the most popular languages being English, German and Russian, in that order.

https://ec.europa.eu/commfrontoffice/publicopinion/archives/...

> relatively unaffected - by the covid19 pandemic

With 2k daily new cases, unaffected my ass.

I used the weasel words "so far". It's more like 2400 at the moment. The fact that the epidemic develops later supports my statement that Poland is a backwater country. Which doesn't mean it will go around us.

https://www.worldometers.info/coronavirus/country/poland/

The R (effective reproduction rate) coefficient is between 1.3 and 1.5 depending on the interval chosen, which is a very worrying trend. But remember Italy was 4.5 in February. Of course, if they cook books to hide financial deficit they sure can fake infection stats.

http://trackingr-env.eba-9muars8y.us-east-2.elasticbeanstalk...

While sampling equally from different income groups might clear up these variables, from what I've been able to find some of highest rates of dementia are in relatively small rich European countries.

Those countries have longer life expectancies so it's hard to do apples-to-apples for rich countries to poor countries.

I do think the "use it or lose it" model for neurodegeneration makes a lot of sense, in the same way you lose muscles if you don't use them, so bilingual is an interesting proxy for greater brain use. Although I also think the lack-of-sleep model makes sense too so I really don't know.

Another from Italy with 85 people that found the same

https://www.alzheimers.org.uk/news/2018-05-15/bilingual-brai...

This one appears to have brain scans to suggest structural differences do exist

N = 266 isn't really that bad at all

What about N = 266, all from one city?

We did FDA cert. of medical device with way less. With sound methodology this number would be totally ok (depending on strength of measured signal)

Reply on news.ycombinator.com