Biomedical research methodology is being dumbed down, making it impossible for research scientists to render results that lead to knowledge and understanding.
Yikes. Now go see if there was an increase in the use of bayesian statistics. It was just getting popular in my dept as I was finishing up back at the turn of the century.
In a different dept, on a different campus, there was a strong reliance on closed source software, magic black boxes, and proprietary data formats. Then they wondered why the people with analytical tools weren't actually interested in collaborating. 🙄
It's flat - and the average is 0.07%. So it's a non-starter. The PIs on research studies are responsible for defining the specific aims, and the NIH has been responsible for setting the direction of research programs. We need wholescale reform.
Doctors around the world came together to sign a Declaration of an International Medical Crisis due to the diseases and death co-related to the COVID-19 “vaccines.”
I'm pretty sure it's much worse than this, because if they've completely junked this kind of analysis, then one can be sure that they've junked rigour in their work in general. It's the new procrustean research methodology of making the facts fit the model, in all respects.
As unsettling as this is, I am grateful for you uncovering poor methodology as a standard practice. If we can't identify the problem, we can't tackle the solution. Is there any policy in universities or the NIH, etc. which has bare minimum standards to approve methodology prior to starting research?
I'd rather create 80 independent research institutions funded by the public #PlanB - at least one in every state - disconnected from profit motives. Search for #PlanB with my name.
“This means the expenditure on research by Congress has become a calamitous waste.” —-It’s not a waste to the mouth-breathing politicians whose North Star is vote-buying and who consider facts to be inconvenient to whatever narratives they’re pushing at the moment to cement their own power.
I appreciated this post a lot, and didn't realize there had been this decline. As an MS biostats who does occasional consulting work still, I can say there has been a big increase in complex modeling but there is less and less simply looking at the data at basic levels, checking assumptions, checking for interactions, etc. I came into a project once and was asked to re-analyze some data where there turned out to be an enormous outlier in the main predictor which had driven the earlier results. Without this single value one reached the opposite conclusion. This analysis had gotten to the manuscript phase with several pairs of eyes on it and no one had apparently done so much as a scatter plot, or they would have seen the issue. Amazing. This is at a globally respected university.
In cooking, quality ingredients and basic things done well are more important than the fanciest techniques or presentations. This is true of statistics too. Far too little thought is put into gathering good, high quality data and then simply getting to know these data properly, without the black box models du jour.
You can't be much of a MS biostatist - you didn't even questions the results or look at the source data for this one. Completely unreproducible using the stated data sources. As you said
"Far too little thought is put into gathering good, high quality data and then simply getting to know these data properly"
Because RCTs are true experiments that lend themselves to proper design to allow testing for interactions - and the inference of causality is built-in.
I didn't realize you had restricted your original searches to RCTs, and thought that this was a point of divergence in the commenter's searches that tried to reproduce yours.
Looks like you have done nothing but expose your bias. If you drop the quotation marks, and do the searches, you will find that the results ARE reproducible.
No worries, though, I don't hold a grudge for your presumptuous judgement, and take full responsibility for being inexact on the use of quotation. You've raised a valid point about being precise in methods descriptions, I concur - but then what's more correct, with, or without quotes? One can do any number of such searches: multiple linear regression or multiple linear regression, or regression, with or without quotes. Since authors use various means of referencing their use of regression, an argument could be made for lexical flexibility to not use quotes. I guess a larger, formal natural language processing analysis would be called for. I have colleagues who are expert in that area. I don't think trends in statistical methodology are sufficiently studied. My original analysis still has me concerned.
No - you exposed your sloppy experimental analysis and biased bombast. You can't tell me that you didn't notice all the papers that never should have been included in your totals - a sampling for the first 25 results:
multiple linear regression analysis. x 2 - IN
multiple sclerosis x 5 (with regression buried somewhere else) - OUT
multiple myeloma (RRMM) x 3 - (with regression buried somewhere else) - OUT
multiple Cox regression analyses (not the same as multiple regression) - OUT
Multiple ordered logistic regression (not the same as multiple regression)
multiple regression analysis - IN
multiple exercise (with regression buried somewhere else) - OUT
multiple clearance estimates (with regression buried somewhere else) - OUT
multiple micronutrients / logistic regression - OUT
multiple correlation analysis and binary logistic regression models - OUT
multiple regression - IN
Sadly, you were worse than imprecise. Your article makes a bombastic claim that is backed by poorly researched numbers. And if I hadn't called you out on your mistakes, you wouldn't have corrected. you dreck doesn't stand up to peer review and that's why you hide it on Substack, where you only get scrutinized by yes-men.
Nice, so that is 8 out 96 total that include "multiple sclerosis" AND "regression". And guess what - my query found all 8 of those as well. 88 of them are likely not applicable (funny that you see that as a win). But glad you are finally learning how these queries work.
It's funny - your commentary and disagreement with me is proof that I only get feedback from yes-men. I prefer OPEN rational discourse and debate. There's nothing stopping me from doing a more intensive analysis and publishing the results in a peer-reviewed journal, if I chose to.
The general argument that you raise and explore is extremely important, has a history of expressed concern and will hopefully draw further constructive interest, suggestions and critique. Keep open to the concerns and ideas proposed, set aside any negative tone that might be disturbing or distracting.
This is a serious issue. Perhaps you can claim bombast by me, because I see and call out the decay. This article might help you put the problem into a less... explicit context, which you apparently need. See "The Effect of Ignoring Statistical Interactions in Regression Analyses Conducted in Epidemiologic Studies: An Example with Survival Analysis Using Cox Proportional Hazards Regression Model" published in 2016 by Kristina Vatcheva at The University of Texas Rio Grande Valley, Joseph Mccormick at the University of Texas Health at Houston and Mohammad Rahbar. Here's their abstract Objective: To demonstrate the adverse impact of ignoring statistical interactions in regression models used in epidemiologic studies. Study design and setting: Based on different scenarios that involved known values for coefficient of the interaction term in Cox regression models we generated 1000 samples of size 600 each. The simulated samples and a real life data set from the Cameron County Hispanic Cohort were used to evaluate the effect of ignoring statistical interactions in these models. Results: Compared to correctly specified Cox regression models with interaction terms, misspecified models without interaction terms resulted in up to 8.95 fold bias in estimated regression coefficients. Whereas when data were generated from a perfect additive Cox proportional hazards regression model the inclusion of the interaction between the two covariates resulted in only 2% estimated bias in main effect regression coefficients estimates, but did not alter the main findings of no significant interactions. Conclusions: When the effects are synergic, the failure to account for an interaction effect could lead to bias and misinterpretation of the results, and in some instances to incorrect policy decisions. Best practices in regression analysis must include identification of interactions, including for analysis of data from epidemiologic studies.
Your own analysis shows that perhaps as many as 95% of studies ignore interaction terms (subject to refinement, of course): here's the link for you.
Multiple Cox Regression also allows the study of interaction terms, so lexical flexibility is less constrained. I would think the most representative search would be inclusive of all regression methods.
I'm amazed at how many dupes took this assessment at face value instead of (easily) running the numbers themselves. Unless the author's methodology was far different than what I am doing below, the author cooked the numbers to feed your biases (exactly what many of you asserted researchers are doing).
To find out, I searched Pubmed for research articles that used the term “multiple regression”
I then calculated for each year the percentage of studies published in Pubmed that used multiple regression and considered the interaction of variables.
Download the two spreadsheets, combine the results and compute the percentage and you get this yearly data (pretty much a flat line between 3 and 5%)
So sorry, I believe I restricted my analysis to clinical trials. I'm used to thinking of "studies" as actual research studies, not reviews, systematic reviews, etc.
My numbers were far lower than yours.
"Multiple regression" w/ "interaction"
2022 1
2021 5
2020 7
2019 2
2018 3
2017 2
2016 4
2015 1
2014 1
2013 1
2012 3
2011 3
2010 1
2009 2
2008 1
Multiple regression w/o interaction
2022 131
2021 208
2020 128
2019 98
2018 91
2017 64
2016 64
2015 57
2014 32
2013 35
2012 34
2011 25
2010 21
2009 14
2008 9
I am not interested in feeding biases. I'm interested in changing Science away from reliance on biases. There are many other trends happening; CDC will now be relying on non-reviewed preprints more in the formation of policy, creating policy before science has had time to validate with replication and meta-analyses. NIH researchers' demands on our national budget are not subject to extramural review. This creates waste and prevents the best research from going forward at Universities. Corporations endow departments w/lifetime Chairs - and threaten to cancel funding if faculty members publish findings against their products. Given this track record, more funds for this kind of BS will create more harm.
Maybe I'm doing something differently, but when I restrict the results to just Clinical Trials I get a very different set of numbers - smaller yes, but both numerator and denominator are smaller. And to be clear, I'm using the count of all Clinical Trial papers with "multiple regression" vs. the number that also include "interaction". Limiting to Clinical Trials still gives me results far from those you highlighted (see below)
It might help ig we compared links to get results:
Total Clinical Trials that include "multiple regression"
The main point is too few studies are designed to consider interactions - and if you actually read how retrospective vaccine studies conducted by CDC or their contractees have been conducted, you will see that their M.O. - codified in a CDC "White Paper" - is to "adjust for" "confounders" - NOT look at interactions between age, birthweight, mother's income, gestational age & vaccine exposure. Unless they look at & publish the interaction terms, the alleged loss of significance of vaccines as a factor is a logic error in analysis. Yet that's their M.O.
Sorry, this is all just obfuscation of the fact that you are purveying incorrect numbers. What is most troubling is that you are showing MORE Clinical Studies papers that include "multiple regression" than actually exist in the most up-to-date queries. That's clearly impossible - you are using provably fallacious data to validate your bias expressed in this previous comment.
The trend is completely opposite what you describe. Charted here with sources. Not sure why you can't give me URL links for the exact filter settings you used. I have included both link in my chart showing your results is impossible to replicate with said data.
Let me try to exactly replicate your results using a moving-target database and then accuse you of obfuscation. I'll do a live video this week, recording each step of my analysis this week and then you can retract your defamatory language. Did you replicate the negative trend, or not?
There are NO meaningful details here that would allow me to reproduce your alleged data. The data that I do extract gives a very different set of numbers and results. I put your numbers in the unsubstantiated category.
Use "Multiple regression" searching only Randomized Clinical trials.
Search between 2008 and 2022
Use "Multiple regression" and "interaction" searching only Randomized Clinical trials.
Proceed from there.
Those results will show a similar trend, but the numbers will change from day to day as journals are added to Pubmed and as new journal articles are loaded for already-Pubmed listed journals.
Your own results using more general search will also change from week to week.
Here's what the claimed data actually looks like when charted. Percentage including "interaction" is generally increasing over the years quoted, not decreasing !
All you need to do is click the two links and download the associated data. Or you can give me the links when you get to the filter settings you were using - much better than video.
That's great - thanks for figuring it out - I worked really hard to be clear on the query I was doing because in query land multiple regression and "multiple regression" are two very different searches. The first finds studies that use "multiple" AND "regression" anywhere in the papers. The second only finds the studies where those two words were used together.
Based on what I'm seeing triggered by your search is a mixed bag:
multiple linear regression analysis. - IN
multiple sclerosis fatigue (with regression buried somewhere else) - OUT
multiple Cox regression analyses (not the same as multiple regression) - OUT
Multiple ordered logistic regression (not the same as multiple regression)
multiple regression analysis - OUT
multiple exercise (with regression buried somewhere else) - OUT
multiple clearance estimates (with regression buried somewhere else) - OUT
multiple micronutrients / logistic regression - OUT
multiple correlation analysis and binary logistic regression models - OUT
multiple regression - IN
So clearly not all related to multiple regression, but some are, mostly the ones that include "multiple linear regression"
The right thing to do would be to do a query that ORs "multiple regression" with "multiple linear regression", then the same query that also ANDs "interaction". I'll do it a little later tonight.
Did you consider the interaction term involving Pharma money and researchers?
If researchers' salaries are somehow tied to useful knowledge produced, this would not be a problem.
Today there is a disincentive to understand root cause of ANY problem. If someone found the root cause of autism today, billions in research dollars and jobs would evaporate. So no one wants to find the root cause of ANYTHING. If you don't have to produce results, why do the methods matter?
There is that... and you know I consider that. Making significant beta coefficients not significant by adding variables that bury their effect in the interaction and not studying it is likely to prove to be more widespread form of research fraud than we suspected given how expert CDC and its contractees have been w/vaccines. Academics have to play nice about it... but a systematic analysis or ten of specific questions about research on corporate products funded by corporations would be useful.
With enough money Pharma can and does buy whatever answer it needs! How else do you think they became the leading cause of death? I know that retired editors from JAMA & NJM etc. claimed over 50% of drug & medical studies were fraudulent! (That 5 billion in fraud fines substantiated that!) I just love that, no jail time, for the perps who caused mass deaths! I.E. stuff “brave” intellects & people don’t talk about, for fear of Ridicule!
I hope not. See http://ipaknowledge.org and let's fund independent research that WILL consider interactions. I want to hire displaced academics
Hi, would there be job offers for people like post docs?
After the pandemic behavior we witnessed they are no longer worthy of the term researchers they are politicized trash.
far too many people have yet to realize this
Sharing, I appreciate what you are doing James Lyons-Weiler.
Yikes. Now go see if there was an increase in the use of bayesian statistics. It was just getting popular in my dept as I was finishing up back at the turn of the century.
In a different dept, on a different campus, there was a strong reliance on closed source software, magic black boxes, and proprietary data formats. Then they wondered why the people with analytical tools weren't actually interested in collaborating. 🙄
It's flat - and the average is 0.07%. So it's a non-starter. The PIs on research studies are responsible for defining the specific aims, and the NIH has been responsible for setting the direction of research programs. We need wholescale reform.
Medical Crisis Declaration
Doctors around the world came together to sign a Declaration of an International Medical Crisis due to the diseases and death co-related to the COVID-19 “vaccines.”
https://jamesroguski.substack.com/p/medical-crisis-declaration
Transhumanism, Nanotechnology and Cybernetics
https://lionessofjudah.substack.com/p/transhumanism-nanotechnology-and
I'm pretty sure it's much worse than this, because if they've completely junked this kind of analysis, then one can be sure that they've junked rigour in their work in general. It's the new procrustean research methodology of making the facts fit the model, in all respects.
As unsettling as this is, I am grateful for you uncovering poor methodology as a standard practice. If we can't identify the problem, we can't tackle the solution. Is there any policy in universities or the NIH, etc. which has bare minimum standards to approve methodology prior to starting research?
I'd rather create 80 independent research institutions funded by the public #PlanB - at least one in every state - disconnected from profit motives. Search for #PlanB with my name.
“This means the expenditure on research by Congress has become a calamitous waste.” —-It’s not a waste to the mouth-breathing politicians whose North Star is vote-buying and who consider facts to be inconvenient to whatever narratives they’re pushing at the moment to cement their own power.
Yep - funding for science has become politicized bacon brought home. That's why we need #PlanB.
I appreciated this post a lot, and didn't realize there had been this decline. As an MS biostats who does occasional consulting work still, I can say there has been a big increase in complex modeling but there is less and less simply looking at the data at basic levels, checking assumptions, checking for interactions, etc. I came into a project once and was asked to re-analyze some data where there turned out to be an enormous outlier in the main predictor which had driven the earlier results. Without this single value one reached the opposite conclusion. This analysis had gotten to the manuscript phase with several pairs of eyes on it and no one had apparently done so much as a scatter plot, or they would have seen the issue. Amazing. This is at a globally respected university.
In cooking, quality ingredients and basic things done well are more important than the fanciest techniques or presentations. This is true of statistics too. Far too little thought is put into gathering good, high quality data and then simply getting to know these data properly, without the black box models du jour.
Please email me, I may need your collaboration/consulting some time!
Thanks, that would be great actually! Is the contact page at IPAK best?
I would try that, yes.
You can't be much of a MS biostatist - you didn't even questions the results or look at the source data for this one. Completely unreproducible using the stated data sources. As you said
"Far too little thought is put into gathering good, high quality data and then simply getting to know these data properly"
That's what this author did.
https://davidmuncier.substack.com/p/incorrect-analysis
We'll see about that.
Why on earth would you look at RCTs only?
Because RCTs are true experiments that lend themselves to proper design to allow testing for interactions - and the inference of causality is built-in.
I didn't realize you had restricted your original searches to RCTs, and thought that this was a point of divergence in the commenter's searches that tried to reproduce yours.
Apologies for that oversight!
"Because RCTs are true experiments that lend themselves to proper design"
Otoh, RCTs also lend themselves to improper design, depending on the motivation of the designer.
"For every muslim who's not a Christian, I can find a Christian who's not a Christian."
Looks like you have done nothing but expose your bias. If you drop the quotation marks, and do the searches, you will find that the results ARE reproducible.
No worries, though, I don't hold a grudge for your presumptuous judgement, and take full responsibility for being inexact on the use of quotation. You've raised a valid point about being precise in methods descriptions, I concur - but then what's more correct, with, or without quotes? One can do any number of such searches: multiple linear regression or multiple linear regression, or regression, with or without quotes. Since authors use various means of referencing their use of regression, an argument could be made for lexical flexibility to not use quotes. I guess a larger, formal natural language processing analysis would be called for. I have colleagues who are expert in that area. I don't think trends in statistical methodology are sufficiently studied. My original analysis still has me concerned.
No - you exposed your sloppy experimental analysis and biased bombast. You can't tell me that you didn't notice all the papers that never should have been included in your totals - a sampling for the first 25 results:
multiple linear regression analysis. x 2 - IN
multiple sclerosis x 5 (with regression buried somewhere else) - OUT
multiple myeloma (RRMM) x 3 - (with regression buried somewhere else) - OUT
multiple Cox regression analyses (not the same as multiple regression) - OUT
Multiple ordered logistic regression (not the same as multiple regression)
multiple regression analysis - IN
multiple exercise (with regression buried somewhere else) - OUT
multiple clearance estimates (with regression buried somewhere else) - OUT
multiple micronutrients / logistic regression - OUT
multiple correlation analysis and binary logistic regression models - OUT
multiple regression - IN
Sadly, you were worse than imprecise. Your article makes a bombastic claim that is backed by poorly researched numbers. And if I hadn't called you out on your mistakes, you wouldn't have corrected. you dreck doesn't stand up to peer review and that's why you hide it on Substack, where you only get scrutinized by yes-men.
There you go again, David. You'll have to go a lot further to convince me that all articles that mention "regression" and happen to also use "multiple" should be ruled out... 8 results just for multiple regression and multiple sclerosis - https://pubmed.ncbi.nlm.nih.gov/?term=%22multiple+regression%22+AND+%22multiple+sclerosis%22&filter=pubt.randomizedcontrolledtrial&filter=dates.2008-2022
Nice, so that is 8 out 96 total that include "multiple sclerosis" AND "regression". And guess what - my query found all 8 of those as well. 88 of them are likely not applicable (funny that you see that as a win). But glad you are finally learning how these queries work.
https://pubmed.ncbi.nlm.nih.gov/?term=%22multiple+sclerosis%22+regression&filter=pubt.randomizedcontrolledtrial&filter=dates.2008-2022
Now try these two:
("multiple regression" OR "multiple linear regression") AND "interaction"
"multiple regression" OR "multiple linear regression"
Here's what you get:
Search query: "multiple regression" OR "multiple linear regression"
Year Count w "interaction" Percentage
2022 76 3 4%
2021 121 8 7%
2020 118 9 8%
2019 117 7 6%
2018 124 7 6%
2017 135 4 3%
2016 125 15 12%
2015 141 8 6%
2014 131 7 5%
2013 149 11 7%
2012 137 6 4%
2011 110 5 5%
2010 108 2 2%
2009 106 4 4%
2008 100 4 4%
"Finally learned" - that's rich. Such a small percentage of studies that might be considering the interaction term. You keep validating my concern.
It's funny - your commentary and disagreement with me is proof that I only get feedback from yes-men. I prefer OPEN rational discourse and debate. There's nothing stopping me from doing a more intensive analysis and publishing the results in a peer-reviewed journal, if I chose to.
I commented early, BTW, not a man.
The general argument that you raise and explore is extremely important, has a history of expressed concern and will hopefully draw further constructive interest, suggestions and critique. Keep open to the concerns and ideas proposed, set aside any negative tone that might be disturbing or distracting.
This is a serious issue. Perhaps you can claim bombast by me, because I see and call out the decay. This article might help you put the problem into a less... explicit context, which you apparently need. See "The Effect of Ignoring Statistical Interactions in Regression Analyses Conducted in Epidemiologic Studies: An Example with Survival Analysis Using Cox Proportional Hazards Regression Model" published in 2016 by Kristina Vatcheva at The University of Texas Rio Grande Valley, Joseph Mccormick at the University of Texas Health at Houston and Mohammad Rahbar. Here's their abstract Objective: To demonstrate the adverse impact of ignoring statistical interactions in regression models used in epidemiologic studies. Study design and setting: Based on different scenarios that involved known values for coefficient of the interaction term in Cox regression models we generated 1000 samples of size 600 each. The simulated samples and a real life data set from the Cameron County Hispanic Cohort were used to evaluate the effect of ignoring statistical interactions in these models. Results: Compared to correctly specified Cox regression models with interaction terms, misspecified models without interaction terms resulted in up to 8.95 fold bias in estimated regression coefficients. Whereas when data were generated from a perfect additive Cox proportional hazards regression model the inclusion of the interaction between the two covariates resulted in only 2% estimated bias in main effect regression coefficients estimates, but did not alter the main findings of no significant interactions. Conclusions: When the effects are synergic, the failure to account for an interaction effect could lead to bias and misinterpretation of the results, and in some instances to incorrect policy decisions. Best practices in regression analysis must include identification of interactions, including for analysis of data from epidemiologic studies.
Your own analysis shows that perhaps as many as 95% of studies ignore interaction terms (subject to refinement, of course): here's the link for you.
https://scholarworks.utrgv.edu/mss_fac/101/
As with Big Pharma's 5 Billion in fraud fines, mostly in riged studies! How Ya gunna believe anything the *leading cause of death cartell tells you?
*see Death by Medicine, Dr. Null.
Multiple Cox Regression also allows the study of interaction terms, so lexical flexibility is less constrained. I would think the most representative search would be inclusive of all regression methods.
Academia is dead. Institutions in general are dead. The individual is the way forward.
I'm amazed at how many dupes took this assessment at face value instead of (easily) running the numbers themselves. Unless the author's methodology was far different than what I am doing below, the author cooked the numbers to feed your biases (exactly what many of you asserted researchers are doing).
To find out, I searched Pubmed for research articles that used the term “multiple regression”
https://pubmed.ncbi.nlm.nih.gov/?term=%22multiple+regression%22
and then I searched for research articles that used the terms “multiple regression” and “interaction”.
https://pubmed.ncbi.nlm.nih.gov/?term=%22multiple+regression%22+%22interaction%22
I then calculated for each year the percentage of studies published in Pubmed that used multiple regression and considered the interaction of variables.
Download the two spreadsheets, combine the results and compute the percentage and you get this yearly data (pretty much a flat line between 3 and 5%)
Search query: "multiple regression"
Year wo interaction with interaction
2022 2741 122 4%
2021 3883 202 5%
2020 3186 160 5%
2019 2868 144 5%
2018 2876 142 5%
2017 2666 122 5%
2016 2739 140 5%
2015 2787 114 4%
2014 2639 113 4%
2013 2613 132 5%
2012 2435 106 4%
2011 2237 95 4%
2010 2013 82 4%
2009 1925 71 4%
2008 1944 66 3%
2007 1740 71 4%
2006 1597 77 5%
2005 1517 62 4%
2004 1341 45 3%
2003 1358 42 3%
2002 1211 38 3%
2001 1176 50 4%
2000 1072 38 4%
1999 1053 47 4%
1998 961 32 3%
1997 957 37 4%
1996 925 29 3%
1995 799 21 3%
1994 727 27 4%
1993 717 21 3%
1992 641 26 4%
1991 542 24 4%
1990 599 24 4%
1989 520 21 4%
1988 388 15 4%
1987 394 21 5%
1986 306 15 5%
1985 303 13 4%
1984 249 16 6%
1983 250 16 6%
1982 227 14 6%
1981 175 10 6%
1980 164 4 2%
1979 136 6 4%
1978 114 3 3%
1977 100 4 4%
1976 93 7 8%
1975 73 3 4%
1974 13 3 23%
1973 22 0%
1972 10 0%
1971 15 0%
1970 10 0%
1969 5 0%
1968 6 0%
1967 9 0%
1966 2 0%
1965 2 0%
1963 3 0%
1962 3 0%
1961 1 0%
1960 1 0%
1959 1 0%
1948 2 0%
So sorry, I believe I restricted my analysis to clinical trials. I'm used to thinking of "studies" as actual research studies, not reviews, systematic reviews, etc.
My numbers were far lower than yours.
"Multiple regression" w/ "interaction"
2022 1
2021 5
2020 7
2019 2
2018 3
2017 2
2016 4
2015 1
2014 1
2013 1
2012 3
2011 3
2010 1
2009 2
2008 1
Multiple regression w/o interaction
2022 131
2021 208
2020 128
2019 98
2018 91
2017 64
2016 64
2015 57
2014 32
2013 35
2012 34
2011 25
2010 21
2009 14
2008 9
I am not interested in feeding biases. I'm interested in changing Science away from reliance on biases. There are many other trends happening; CDC will now be relying on non-reviewed preprints more in the formation of policy, creating policy before science has had time to validate with replication and meta-analyses. NIH researchers' demands on our national budget are not subject to extramural review. This creates waste and prevents the best research from going forward at Universities. Corporations endow departments w/lifetime Chairs - and threaten to cancel funding if faculty members publish findings against their products. Given this track record, more funds for this kind of BS will create more harm.
Maybe I'm doing something differently, but when I restrict the results to just Clinical Trials I get a very different set of numbers - smaller yes, but both numerator and denominator are smaller. And to be clear, I'm using the count of all Clinical Trial papers with "multiple regression" vs. the number that also include "interaction". Limiting to Clinical Trials still gives me results far from those you highlighted (see below)
It might help ig we compared links to get results:
Total Clinical Trials that include "multiple regression"
https://pubmed.ncbi.nlm.nih.gov/?term=%22multiple+regression%22+&filter=pubt.clinicaltrial
Clinical Trials that include "multiple regression" and "interaction"
https://pubmed.ncbi.nlm.nih.gov/?term=%22multiple+regression%22+%22interaction%22&filter=pubt.clinicaltrial
Search query: "multiple regression"
Year total w interaction percentage
2022 55 2 4%
2021 89 7 8%
2020 89 8 9%
2019 87 6 7%
2018 114 5 4%
2017 117 2 2%
2016 120 10 8%
2015 139 7 5%
2014 113 6 5%
2013 130 10 8%
2012 127 9 7%
2011 122 2 2%
2010 111 1 1%
2009 128 2 2%
2008 113 4 4%
2007 108 4 4%
2006 91 3 3%
2005 113 10 9%
2004 103 6 6%
2003 99 3 3%
2002 92 4 4%
2001 74 4 5%
2000 90 3 3%
1999 112 5 4%
1998 98 7 7%
1997 85 3 4%
1996 71 5 7%
1995 79 2 3%
1994 67 5 7%
1993 34 2 6%
1992 29 1 3%
1991 24 1 4%
1990 25 1 4%
1989 12 1 8%
1988 12 1 8%
1987 8 0 0%
1986 9 0 0%
1985 10 0 0%
1984 8 0 0%
1983 3 0 0%
1982 9 0 0%
1981 6 0 0%
1979 2 0 0%
1978 2 0 0%
1977 6 0 0%
1975 1 0 0%
1974 1 0 0%
1971 1 0 0%
Pubmed is constantly updated, a moving target.
The main point is too few studies are designed to consider interactions - and if you actually read how retrospective vaccine studies conducted by CDC or their contractees have been conducted, you will see that their M.O. - codified in a CDC "White Paper" - is to "adjust for" "confounders" - NOT look at interactions between age, birthweight, mother's income, gestational age & vaccine exposure. Unless they look at & publish the interaction terms, the alleged loss of significance of vaccines as a factor is a logic error in analysis. Yet that's their M.O.
Sorry, this is all just obfuscation of the fact that you are purveying incorrect numbers. What is most troubling is that you are showing MORE Clinical Studies papers that include "multiple regression" than actually exist in the most up-to-date queries. That's clearly impossible - you are using provably fallacious data to validate your bias expressed in this previous comment.
No comment on the trend you found, eh?
The trend is completely opposite what you describe. Charted here with sources. Not sure why you can't give me URL links for the exact filter settings you used. I have included both link in my chart showing your results is impossible to replicate with said data.
https://davidmuncier.substack.com/p/incorrect-analysis
Let me try to exactly replicate your results using a moving-target database and then accuse you of obfuscation. I'll do a live video this week, recording each step of my analysis this week and then you can retract your defamatory language. Did you replicate the negative trend, or not?
The trend is completely opposite what you describe. Charted here with sources.
https://davidmuncier.substack.com/p/incorrect-analysis
See more detail here: https://bit.ly/3B7xsMG
There are NO meaningful details here that would allow me to reproduce your alleged data. The data that I do extract gives a very different set of numbers and results. I put your numbers in the unsubstantiated category.
Search between 2008 and 2022
Use "Multiple regression" searching only Randomized Clinical trials.
Search between 2008 and 2022
Use "Multiple regression" and "interaction" searching only Randomized Clinical trials.
Proceed from there.
Those results will show a similar trend, but the numbers will change from day to day as journals are added to Pubmed and as new journal articles are loaded for already-Pubmed listed journals.
Your own results using more general search will also change from week to week.
Here's the link to your first exact query:
https://pubmed.ncbi.nlm.nih.gov/?term=%22multiple+regression%22&filter=pubt.randomizedcontrolledtrial&filter=dates.2008-2022
Results are below. Just to remind you, you are claiming 131 papers for 2022. Try for yourself.
Search query: "multiple regression"
Year Count
2022 37
2021 63
2020 64
2019 56
2018 70
2017 82
2016 69
2015 91
2014 68
2013 91
2012 78
2011 71
2010 72
2009 78
2008 74
Here's what the claimed data actually looks like when charted. Percentage including "interaction" is generally increasing over the years quoted, not decreasing !
https://davidmuncier.substack.com/p/incorrect-analysis
All sources are URL linked so you can download the data yourself and do the same.
We'll see if I can replicate YOUR results with the searches I do on video.
All you need to do is click the two links and download the associated data. Or you can give me the links when you get to the filter settings you were using - much better than video.
It was the quotes, see my other comments. My results, it turns out, are easily reproduced.
That's great - thanks for figuring it out - I worked really hard to be clear on the query I was doing because in query land multiple regression and "multiple regression" are two very different searches. The first finds studies that use "multiple" AND "regression" anywhere in the papers. The second only finds the studies where those two words were used together.
Based on what I'm seeing triggered by your search is a mixed bag:
multiple linear regression analysis. - IN
multiple sclerosis fatigue (with regression buried somewhere else) - OUT
multiple myeloma (RRMM) - (with regression buried somewhere else) - OUT
multiple Cox regression analyses (not the same as multiple regression) - OUT
Multiple ordered logistic regression (not the same as multiple regression)
multiple regression analysis - OUT
multiple exercise (with regression buried somewhere else) - OUT
multiple clearance estimates (with regression buried somewhere else) - OUT
multiple micronutrients / logistic regression - OUT
multiple correlation analysis and binary logistic regression models - OUT
multiple regression - IN
So clearly not all related to multiple regression, but some are, mostly the ones that include "multiple linear regression"
The right thing to do would be to do a query that ORs "multiple regression" with "multiple linear regression", then the same query that also ANDs "interaction". I'll do it a little later tonight.
Did you consider the interaction term involving Pharma money and researchers?
If researchers' salaries are somehow tied to useful knowledge produced, this would not be a problem.
Today there is a disincentive to understand root cause of ANY problem. If someone found the root cause of autism today, billions in research dollars and jobs would evaporate. So no one wants to find the root cause of ANYTHING. If you don't have to produce results, why do the methods matter?
There is that... and you know I consider that. Making significant beta coefficients not significant by adding variables that bury their effect in the interaction and not studying it is likely to prove to be more widespread form of research fraud than we suspected given how expert CDC and its contractees have been w/vaccines. Academics have to play nice about it... but a systematic analysis or ten of specific questions about research on corporate products funded by corporations would be useful.
With enough money Pharma can and does buy whatever answer it needs! How else do you think they became the leading cause of death? I know that retired editors from JAMA & NJM etc. claimed over 50% of drug & medical studies were fraudulent! (That 5 billion in fraud fines substantiated that!) I just love that, no jail time, for the perps who caused mass deaths! I.E. stuff “brave” intellects & people don’t talk about, for fear of Ridicule!
I just love this. Main effects that are sizeable or meaningful are rare indeed.