A crumb of a clue on epidemiology

[forestglip]

As @Yann mentioned the graphs sometimes seem to some large extent be driven by things related to public awareness which might work in all sorts of unusual ways

I agree with you and Yann, and I think the most likely reason for the correlation is for some reason more awareness of ME/CFS in people with British ancestry. But I think it's a fun idea that has a small chance of actually being about prevalence.

 

[EndME]

I agree with you and Yann, and I think the most likely reason for the correlation is for some reason more awareness of ME/CFS in people with British ancestry. But I think it's a fun idea that has a small chance of actually being about prevalence.

Maybe it's all the Guardian and George Monbiots fault after all! I agree nothing is lost with playing around with this as @Murph suggested.

 

[forestglip]

The census data includes 108 ancestries. I ran the regression on all of them against the same Google Trends values. Here are the top 20 highest R^2 values:

This was showing lots of UK countries at the top, and I realized that it may be because people could report multiple ancestries.

From an overview of these datasets:

The ACS asks each respondent to write their ancestry or ethnic origin, and records up to two ancestries per person (the first two ancestries written by the respondent).

The table for People Reporting Single Ancestry (B04004) shows data for those who reported only one ancestry, while People Reporting Multiple Ancestry (B04005) shows data for those who reported more than one ancestry. People Reporting Ancestry (B04006) shows data for those who reported any ancestry, regardless of whether it was the only ancestry or part of multiple ancestries they reported.

Note: this means that values in B04005 and B04006 will not necessarily add up to match totals, because one person may be represented under more than one ancestry.

I (and I think Murph) used B04006, which counts up to two ancestries per person. So, for example, the correlations for Scottish and English might both be high because the same people reported both ancestries.

I tried again with B04004, which only includes people who reported one ancestry, to avoid double counting and allow better comparison between ancestries.

In this case, the correlation with English is pretty much gone (the sample size is also much smaller for this dataset, so the ancestry values may be less precise):


There are still some large correlations in this analysis, and British is still near the top at #9, though less significant, with an R^2=0.26 and p=0.0018.


The top 5 correlations with Google Searches for ME/CFS are Northern European, European, Swedish, Icelander, and New Zealander.

Here for example is the plot of the top correlation, Northern European ancestry vs Google Trends for ME/CFS:


(Note that only 35 states were included as the rest had missing data.)

However, it looks like the correlation might be mainly driven by just the five states in the upper right. If I exclude those, the correlation is much less apparent:


Edit: Added confidence bars to plots.

Edit: Maybe worth checking if the association is due to income level.

 

[Felis Catus]

Are these searches in English only?

 

[forestglip]

Are these searches in English only?

This page seems to suggest that if you are looking at a simple search term, then it will show trends just for that specific term in that language. But if you search by a "Topic", then it tracks trends across different translated versions of that topic.

- https://support.google.com/trends/answer/4359550?hl=en#zippy=,compare-terms-across-languages

if you enter ねこ, the Japanese characters for "cat," you won’t find much data for the US since many people in the US search for "cat."

Topics are a group of terms that share the same concept in any language. You can find topics below your search term.

For example, if you search London, and choose the corresponding topic, your search includes results for topics like "Capital of the UK" and "Londres," which is "London" in Spanish.

So I think the "Myalgic encephalomyelitis/chronic fatigue syndrome" I linked before is a "Topic" that should transcend language.

 

[forestglip]

Hm. I really thought it might be that people with higher income search ME/CFS more, as they might have better access to resources that would make them aware of what ME/CFS is. And maybe states with larger proportions of Northern European or British ancestry would have higher average income.

But the correlation is very weak between search trends for ME/CFS and median household income in each state.

This is again the Trends data I used in post #4: https://trends.google.com/trends/explore?q=/m/0dctd&geo=US&legacy&hl=en

And I used income data from the following dataset, using the "Households" row, and the "Median income (dollars)" column for each state: https://data.census.gov/table/ACSST...ag-Grid-AutoColumn~(Margin+of+Errorundefined)



Controlling for Income in the regression of search trend vs. English ancestry, ancestry still significantly predicts the trend value.


Edit: Replaced model summary with the regression results using original unscaled values. I had originally used scaled values because I thought it was causing issues with the regression due to income and ancestry being around 7 orders of magnitude apart, but the results are the same.

"Households" is the income variable.

 

[Utsikt]

A Norwegian study found that higher education was correlated with getting an ME/CFS diagnosis, but lower income was also correlated.

It was speculated that high education = more access to knowledge and ability to pursue better medical care, and that lower income = higher need for a diagnosis for benefit purposes etc.

 

[forestglip]

A Norwegian study found that higher education was correlated with getting an ME/CFS diagnosis, but lower income was also correlated.

It was speculated that high education = more access to knowledge and ability to pursue better medical care, and that lower income = higher need for a diagnosis for benefit purposes etc.

I like your thinking.

I used the 2024 5-year average S1501 "Educational Attainment" dataset, specifically the column for percent of those at least 25 years old that have a bachelor's degree.

Same 5-year English ancestry and 1-year Trends data as before.

There does seem to be a correlation of educational attainment with the Google Trends for ME/CFS:



But it still doesn't make the English ancestry correlation less significant to control for proportion of people who have a bachelor's degree:



I also tried with a 5 year time frame from 2019 to 2024 for the Trends data to try to roughly match the time frames of the 5 year Census datasets I'm using, and both English ancestry and Educational attainment are still significantly associated with searching for ME/CFS.



Edit: Note that the variable name for English is called "Estimate!!Total:!!English" as the dataset has the total number with that ancestry, not a proportion, but I converted it to a proportion of the total population in the state. I probably should change the variable name after converting to make that clear.

 

[Kitty]

both English ancestry and Educational attainment are still significantly associated with searching for ME/CFS.

I wonder if there are associations with searching for 'chronic fatigue syndrome'?

(Sorry, that's an idle thought with no idea of the work involved—I don't understand maths and statistics!)

 

[forestglip]

I wonder if there are associations with searching for 'chronic fatigue syndrome'?

For comparison, here are the plots for association of 5 year search data for the ME/CFS topic with English ancestry and education:


Interestingly, if instead using the specific search term "chronic fatigue syndrome" over the same time frame, the association with English ancestry gets stronger, but the association with education disappears:

 

[Kitty]

Interesting! I suppose the thing we can't know is what people think 'English ancestry' means, and whether they were asked about other ancestries?

 

[forestglip]

whether they were asked about other ancestries?

They asked about many ancestries. I checked the associations with all of them in these two posts:
- https://www.s4me.info/threads/a-crumb-of-a-clue-on-epidemiology.49455/post-682605
- https://www.s4me.info/threads/a-crumb-of-a-clue-on-epidemiology.49455/post-682794

 

[jnmaciuch]

Interesting! I suppose the thing we can't know is what people think 'English ancestry' means, and whether they were asked about other ancestries?

the specific question asked is “what is this person’s ancestry or ethnic origin?” and then it lists several examples like Italian, Jamaican, Korean, etc.

It’s a write-in question so I guess people could technically write anything and then it gets coded by the bureau staff as appropriate. It seems like they code the first two someone lists

Frequently Asked Questions (FAQs) About Ancestry

Browse and explore FAQs on census.gov by topic or group. Click on Submit Request to reach our support team. We will find the answer for you soon. Thank you.

www.census.gov

Thats why im pretty certain people are just guessing based on their last name unless their family recently emigrated. You do have the option of writing “I don’t know” but the question doesnt really make that obvious

 

[Kitty]

Thats why im pretty certain people are just guessing based on their last name unless their family recently emigrated.

That's what I suspected.

In parts of the US, 'English' tends to include Welsh, Scots, Irish and Cornish, which might make no difference at all or add another layer of meaninglessness. It depends how far back you want the definition of ancestry to go, or whether you really mean ancestral origins or cultural origins.

ETA: or the origins you know about, which might only consist of 50%.

 

[forestglip]

You got R^2=0.52. I got R^2=0.31. It looks like the difference may be due to me using newer Google Trends data or a slightly different search term. Vermont is #1 in mine, but #4 in your analysis. But it's still very significant.

I realized the main reason I got a smaller R2 value was because I used the trend values based only on the last year of search data, while Murph used the data based on searches from 2004 to present. From the blog post:

 

[Liface]

How do you explain the Dakotas being at the bottom, considering ME seems to be highly prevalent in Scandinavia?

As an Oregonian, Oregon, Maine, Vermont, etc. are very crunchy blue tribe states where people are more likely to seek out medical labels.

I don't think many people from West Virginia are searching the exact name of the syndrome, but they may be searching "why am I tired all the time"

 

[forestglip]

An AI gave me some more ideas for what may be driving the association: living in a rural area and having access to internet.

So I downloaded data for those two and included them along with education, income, language, and English ancestry into one regression. But alas, English ancestry remains very significant.


Here are all the data sources:

Reporting English ancestry: https://data.census.gov/table/ACSDT5Y2024.B04006?g=010XX00US$0400000
Internet access: https://data.census.gov/table/ACSST5Y2024.S2801?g=010XX00US$0400000
Live in rural area: https://data.census.gov/table/DECENNIALDHC2020.P2?g=010XX00US$0400000
Proportion of >25 year olds with Bachelor's degree: https://data.census.gov/table/ACSST5Y2024.S1501?g=010XX00US$0400000
Median household Income: https://data.census.gov/table/ACSST5Y2024.S1903?g=010XX00US$0400000
Speak only English at home: https://data.census.gov/table/ACSST5Y2024.S1601?g=010XX00US$0400000

Search trends: https://trends.google.com/trends/explore?date=2004-01-01 2026-03-24&geo=US&q=/m/0dctd&hl=en-US

And here is the dataset I used with all 7 of these variables, in case anyone wants to verify:

Spoiler: Dataset

Region,Estimate!!Total:!!English_proportion,Estimate!!Percent!!Total households!!TYPE OF INTERNET SUBSCRIPTIONS!!With an Internet subscription:, !!Total:!!Rural_proportion,Estimate!!Median income (dollars)!!HOUSEHOLD INCOME BY RACE AND HISPANIC OR LATINO ORIGIN OF HOUSEHOLDER!!Households,Estimate!!Percent!!Population 5 years and over!!Speak only English,Estimate!!Percent!!AGE BY EDUCATIONAL ATTAINMENT!!Population 25 years and over!!Bachelor's degree,Myalgic encephalomyelitis/chronic fatigue syndrome: (1/1/04 - 3/24/26)
Alabama,0.12126737449004948,87.7,0.42262760487624196,63999,94.0,17.3,68.0
Alaska,0.09991083394725611,92.1,0.35100512550603974,92788,84.9,19.8,69.0
Arizona,0.10193881475646979,91.7,0.10714840043392283,79964,74.3,20.3,71.0
Arkansas,0.12287797793067534,86.9,0.44523868977965975,60773,91.7,16.3,70.0
California,0.06303223042861833,93.6,0.057633672610931455,99122,55.6,22.7,67.0
Colorado,0.12629156105338807,93.7,0.13973293446817767,95470,83.8,28.1,88.0
Connecticut,0.08685620227628137,92.3,0.13749270648684506,95781,76.6,23.3,75.0
Delaware,0.10472967348909264,92.5,0.17387882999915147,84954,84.9,20.4,63.0
District of Columbia,0.07115723901869091,91.3,0.0,109870,80.9,26.2,68.0
Florida,0.0821181154936254,91.7,0.08465805408783943,74568,69.4,21.3,61.0
Georgia,0.10357064412685926,91.1,0.2593302705736457,77353,84.5,21.1,61.0
Hawaii,0.05125325639082917,92.2,0.1393699180427563,100389,75.3,22.8,62.0
Idaho,0.19265487302133838,92.1,0.3075782472570912,77800,89.0,21.3,80.0
Illinois,0.06855957849821635,90.9,0.1307252256935176,83390,76.0,22.6,67.0
Indiana,0.11289341100291882,90.0,0.28823726023973373,71957,89.7,18.7,79.0
Iowa,0.10210879465455144,89.1,0.36846458826549533,75059,90.9,20.9,69.0
Kansas,0.12424415470021176,90.8,0.27700961237354826,74275,87.9,22.2,69.0
Kentucky,0.14543232548826593,88.8,0.41301547592943905,63726,93.2,16.2,73.0
Louisiana,0.07278509076724629,86.7,0.28458333056018165,60756,91.5,17.0,66.0
Maine,0.21826869104500088,90.6,0.613678186146236,74733,94.0,22.5,94.0
Maryland,0.08457880593508454,92.5,0.14382900798157877,103678,78.9,22.7,75.0
Massachusetts,0.09650278191996202,92.9,0.08720188303787939,103960,74.9,25.5,78.0
Michigan,0.10504148689376538,90.7,0.265256048451718,72875,89.6,19.7,73.0
Minnesota,0.06589922893241419,91.8,0.2812129479151297,89062,87.6,25.6,76.0
Mississippi,0.10595229570999386,84.2,0.5370952888937517,56447,95.7,15.3,64.0
Missouri,0.11861225159541292,89.4,0.3053251930612179,70702,93.3,19.8,71.0
Montana,0.13141309457190822,89.8,0.46581475247296455,72509,95.8,23.2,100.0
Nebraska,0.09411347915583257,90.8,0.2699464288627502,76475,87.2,22.3,66.0
Nevada,0.08665451238643829,92.2,0.05907690940000915,78260,70.5,18.1,65.0
New Hampshire,0.1832144690393643,93.1,0.41676726950938964,99031,92.1,24.5,84.0
New Jersey,0.04910877352052038,92.9,0.06246263050659738,103556,66.9,26.1,72.0
New Mexico,0.08546461118191002,86.7,0.2545286424414953,64059,68.7,17.1,73.0
New York,0.05595322995757785,91.0,0.12554090096112375,85974,69.1,22.4,66.0
North Carolina,0.1274765609943484,90.2,0.332841446261026,72388,86.5,22.2,71.0
North Dakota,0.051291917726010744,88.8,0.3903315902830724,76657,93.3,23.0,61.0
Ohio,0.10986306605602418,90.3,0.23715931457132572,71389,92.0,19.3,68.0
Oklahoma,0.11417815567261756,89.0,0.3537805292935487,65039,88.5,18.5,72.0
Oregon,0.14086805962823906,92.6,0.19500167089267206,83011,84.8,22.7,87.0
Pennsylvania,0.08262084846196288,90.0,0.23546109654148753,77971,87.6,20.9,69.0
Rhode Island,0.10838799383917785,91.9,0.08947501273488923,87796,76.9,22.4,76.0
South Carolina,0.12138287176243456,88.9,0.32051969111591944,69324,91.6,19.8,66.0
South Dakota,0.07678735943788377,89.9,0.42780435044949233,75081,92.9,21.9,62.0
Tennessee,0.13292981713558408,89.3,0.3376663328915154,69595,91.4,19.4,69.0
Texas,0.07888497259744331,91.4,0.1627972478088817,78476,64.9,21.5,59.0
Utah,0.28264105124175676,93.5,0.10218589223185118,95166,84.1,24.7,93.0
Vermont,0.19128705343483138,90.2,0.6487978888997741,81203,94.5,25.2,92.0
Virginia,0.12016835972186643,91.1,0.2436547611723855,93170,82.6,23.7,65.0
Washington,0.12096813813919854,93.7,0.1662815411923329,98141,78.6,24.0,82.0
West Virginia,0.15661000251353344,86.3,0.5535207357240499,59608,97.5,14.4,87.0
Wisconsin,0.06715969508723096,90.6,0.3291686164828382,77485,91.0,22.0,69.0
Wyoming,0.15672470840337432,89.6,0.3798225191600604,76176,93.2,19.1,69.0

There must be some other state-level data that correlates to searches for ME/CFS as well as English and other European ancestries.

I kind of want to test correlation of hundreds or thousands of different state statistics against these ME/CFS searches to try to find something that even has as strong a correlation as English ancestry. Though I'm not sure if it'd be easy to compile lots of datasets to do that.

Edit: Apologies, the original results and dataset I posted were with household income scaled down by 100,000 to test if it changed anything. I've updated both to be the original data. Same conclusions.

 

[forestglip]

How do you explain the Dakotas being at the bottom, considering ME seems to be highly prevalent in Scandinavia?

Good question.

I don't think many people from West Virginia are searching the exact name of the syndrome, but they may be searching "why am I tired all the time"

Just to check how well "tired" and "why am I so tired" correlate to English ancestry, I tested those too with the Google trends data from 2004 to present.

R2=0.11 for "tired" and R2=0.16 for "why am I so tired". So there's a correlation there too, albeit weaker than for ME/CFS.

 

[Nightsong]

I realise there are issues with the dataset which we have discussed before but it might be worth seeing if this correlation holds up in the CDC NHIS dataset. At a brief glance at IPUMS the "Main national origin" (NATORIGIN) field includes "Anglo American" as a possible selection and CFS status is collected (CHRFATIGEV - "Ever had chronic fatigue syndrome" and CFSSTILL - "Still have chronic fatigue syndrome"). Wish I was well enough to help work on this.

 

[forestglip]

it might be worth seeing if this correlation holds up in the CDC NHIS dataset. At a brief glance at IPUMS the "Main national origin" (NATORIGIN) field includes "Anglo American" as a possible selection and CFS status is collected (CHRFATIGEV - "Ever had chronic fatigue syndrome" and CFSSTILL - "Still have chronic fatigue syndrome").

Thanks, the IPUMS website looks very intriguing! Maybe it'll allow me to make a comprehensive datasets of lot of variables to test.

In terms of the variables you mentioned, I think NATORIGIN might have stopped being asked before CHRFATIGEV was started.

NATORIGIN: 1976-1977, 1987, 1992
CHRFATIGEV: 2008, 2021-2024
CFSSTILL: 2021-2024

There are some other ethnicity/nativity variables listed, though none as fine-grained about specific countries as the census. REGIONBR (Global region of birth) might have been good, but it also seems to be unavailable for CHRFATIGEV years. [Edit: Well, there's overlap for one year, 2008.]