about · email me · subscribe

spurious scholar

Because if p < 0.05, why not publish?

Step 1: Gather a bunch of data. There are 25,156 variables in my database. The data ranges from the mundane (air pollution in Chicago) to the weird (Hotdogs consumed by Nathan's Hot Dog Eating Competition Champion) to the super-niche (How clickbait-y Numberphile YouTube video titles are, as rated by an AI).
Step 2: Dredge that data to find random correlations between variables. "Dredging data" means taking one variable and correlating it against every other variable just to see what sticks. It's a dangerous way to go about analysis, because any sufficiently large dataset will yield strong correlations completely at random.

Fun fact: the chart used on the wikipedia page to demonstrate data dredging is also from me. I've been being naughty with data since 2014.

Step 3: Calculate the correlation coefficient, confidence interval, and p-value to see if the connection is statistically significant. "Statistically significant" is a misleading term. It sounds like it means "statistically significant" because, you know, those are the same two words. Unfortunately statistical significance is a technical term that means mumble mumble at least as extreme mumble mumble null hypothesis mumble mumble probability mumble mumble p-values.

You know what? Forget the technical definition. "Statistically significant" just means "someone did some fancy math."

I really did the fancy math below and you can check it by clicking on the "view detailed data" link under each paper. And these really do qualify as "statistically significant" in the technical sense. It's just that "statistically significant" does not mean the results are "significant."

Step 4: If it is, have a large language model draft a research paper.
Step 5: Remind everyone that these papers are AI-generated and are not real. Seriously, just pick one and read the lit review section. The silliness of the papers is an artifact of me (1) having fun and (2) acknowledging that realistic-looking AI-generated noise is a real concern for academic research (peer reviews in particular).

The papers could sound more realistic than they do, but I intentionally prompted the model to write papers that look real but sound silly.

Also: every page says "This paper is AI-generated" at the bottom and the first letters of the names of the authors always spell out C-H-A-T-G-P-T.

Step 6: ...publish:

The Marilyn Effect: A 'Name-worthy' Correlation Between Popularity of the First Name Marilyn and the Number of Economists in Hawaii
The Journal of Quirky Social Science Research
r=0.928 · 95% conf. int. [0.782,0.977] · r2=0.860 · p < 0.01
Generated Jan 2024 · View data details

Joys and Traffic Cohorts: Exploring the Jolly Connection
Journal of Mirthful Mobility Studies
r=0.814 · 95% conf. int. [0.580,0.924] · r2=0.662 · p < 0.01
Generated Jan 2024 · View data details

Ammunition of the Mind: Exploring the Relationship Between Bachelor's Degrees in Military Technologies and Applied Sciences and the Employment of Psychiatric Technicians in Texas
The Journal of Occupational Quirks and Curiosities
r=0.984 · 95% conf. int. [0.930,0.996] · r2=0.968 · p < 0.01
Generated Jan 2024 · View data details

Lingua-Lab Correlations: A Chemical Case Study of Foreign Language Degrees and Wyoming's Plant Operators
The Journal of Multicultural Linguistics and Botanical Engineering
r=0.919 · 95% conf. int. [0.656,0.983] · r2=0.845 · p < 0.01
Generated Jan 2024 · View data details

Airborne Aggravation: Analyzing the Amusing Association Between Air Pollution in Wabash and Customer Contentment with Frontier Communications
Journal of Environmental Humor and Ecological Irony
r=0.788 · 95% conf. int. [0.588,0.897] · r2=0.621 · p < 0.01
Generated Jan 2024 · View data details

A Breath of Fresh Air: Uncovering the Relationship Between Air Quality in Muskegon, Michigan and Forest Cover in the Brazilian Amazon
The Journal of Ecological Connections
r=0.781 · 95% conf. int. [0.609,0.883] · r2=0.611 · p < 0.01
Generated Jan 2024 · View data details

Raw Data: Untangling the Tangled Tale of Sushi and Smog in St. Marys
The Journal of Culinary and Environmental Entanglements
r=0.997 · 95% conf. int. [0.984,0.999] · r2=0.993 · p < 0.01
Generated Jan 2024 · View data details

Just Being Justin: A Breathy Analysis of Air Pollution and the Popularity of the Name Justin in Chicago
The Journal of Urban Air Quality and Sociolinguistics
r=0.764 · 95% conf. int. [0.601,0.866] · r2=0.583 · p < 0.01
Generated Jan 2024 · View data details

The Art of Multiplication: Exploring the Relationship between Master's Degrees in Visual and Performing Arts and the Count of Cashiers in Vermont
Journal of Aesthetic Economics
r=0.985 · 95% conf. int. [0.934,0.996] · r2=0.969 · p < 0.01
Generated Jan 2024 · View data details

Grading the Connection: A Lesson in Correlation Between 2nd Grade Enrollment and SLB's Stock Price
The Journal of Elementary Econometrics
r=0.829 · 95% conf. int. [0.619,0.929] · r2=0.688 · p < 0.01
Generated Jan 2024 · View data details

Masters of Parks, Searching for Narks: The Sushi Near Me Larks
Journal of Culinary Cartography
r=0.976 · 95% conf. int. [0.900,0.995] · r2=0.953 · p < 0.01
Generated Jan 2024 · View data details

A Link Between Sausage and Smog: Assessing the Relationship Between Air Pollution in Port Angeles, Washington and the Consumption of Hotdogs by the Champion of Nathan's Hot Dog Eating Competition
The Journal of Gastronomic Geoscience
r=0.670 · 95% conf. int. [0.438,0.818] · r2=0.449 · p < 0.01
Generated Jan 2024 · View data details

Air Pollution in Flagstaff, Arizona: An Unlikely Muse for the 'Gangnam Style' Craze
The Journal of Eclectic Environmental Studies
r=0.874 · 95% conf. int. [0.602,0.964] · r2=0.764 · p < 0.01
Generated Jan 2024 · View data details

Bakerfield's Bad Air and Brawls: Examining the Link Between Air Pollution and Violent Crime Rates
The Journal of Ecological Criminology
r=0.671 · 95% conf. int. [0.448,0.816] · r2=0.451 · p < 0.01
Generated Jan 2024 · View data details

Jaime's Fame and Reading's Air - A Whiff of Name Popularity and Pollution in Pennsylvania
The Journal of Quirky Sociological Studies
r=0.847 · 95% conf. int. [0.733,0.915] · r2=0.717 · p < 0.01
Generated Jan 2024 · View data details

The Lesley Effect: A Breath of Fresh Air or a Cloud of Pollution?
Journal of Environmental Psychology
r=0.714 · 95% conf. int. [0.527,0.835] · r2=0.510 · p < 0.01
Generated Jan 2024 · View data details

Corny Connections: Genetically Modified Corn in Kansas and the Curious Case of Google Searches for 'Desktop Background'
Journal of Agricultural Genetics and Digital Cultures
r=0.958 · 95% conf. int. [0.884,0.985] · r2=0.917 · p < 0.01
Generated Jan 2024 · View data details

The GMO Conundrum: A Spun Yarn Connecting Genetically Modified Cotton in Arkansas to 'I Can't Even' Google Searches
The Journal of Agricultural Absurdities
r=0.887 · 95% conf. int. [0.724,0.956] · r2=0.787 · p < 0.01
Generated Jan 2024 · View data details

Double Play: The Curious Correlation Between Associates Degrees in Public Administration and Social Services and Baltimore Orioles Wins
Journal of Quirky Social Science
r=0.923 · 95% conf. int. [0.723,0.980] · r2=0.852 · p < 0.01
Generated Jan 2024 · View data details

Management Information Systems: A Gas-tly Connection to Liquefied Petroleum in Lithuania
The International Journal of Quirky Technology Studies
r=0.956 · 95% conf. int. [0.834,0.989] · r2=0.913 · p < 0.01
Generated Jan 2024 · View data details

Degrees of Interest: The Interdisciplinary Connection Between Bachelor's Degrees and Google Search Behavior
The Journal of Interdisciplinary Inquiry and Search Behavior
r=0.971 · 95% conf. int. [0.878,0.993] · r2=0.942 · p < 0.01
Generated Jan 2024 · View data details

Gouda Investment: Unraveling the Ties Between American Cheese Consumption and Ryanair Holdings' Stock Price
Journal of Dairy Economics and Financial Analysis
r=0.937 · 95% conf. int. [0.845,0.975] · r2=0.878 · p < 0.01
Generated Jan 2024 · View data details

Quench the Market: How US Bottled Water Consumption and Northrop Grumman's Stock Price Make Waves
Journal of Market Hydrodynamics
r=0.935 · 95% conf. int. [0.843,0.974] · r2=0.874 · p < 0.01
Generated Jan 2024 · View data details

Meat Your Investments: A Quantitative Study of the Relationship Between Household Spending on Animal Products and Tractor Supply Company's Stock Price
Journal of Agricultural Economics and Financial Markets
r=0.929 · 95% conf. int. [0.831,0.971] · r2=0.864 · p < 0.01
Generated Jan 2024 · View data details

Marching Towards Renewable Energy: Exploring the Link Between Military Technology Master's Degrees and Wind Power in Bosnia and Herzegovina
Journal of Military Technology and Sustainable Energy
r=0.962 · 95% conf. int. [0.843,0.991] · r2=0.926 · p < 0.01
Generated Jan 2024 · View data details


Currently viewing 25 of 4,731 spurious research papers

Page
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190



Why this works

  1. Data dredging: I have 25,156 variables in my database. I compare all these variables against each other to find ones that randomly match up. That's 632,824,336 correlation calculations! This is called “data dredging.” Fun fact: the chart used on the wikipedia page to demonstrate data dredging is also from me. I've been being naughty with data since 2014.
    Instead of starting with a hypothesis and testing it, I isntead tossed a bunch of data in a blender to see what correlations would shake out. It’s a dangerous way to go about analysis, because any sufficiently large dataset will yield strong correlations completely at random.
  2. Lack of causal connection: There is probably no direct connection between these variables, despite what the AI says above. Because these pages are automatically generated, it's possible that the two variables you are viewing are in fact causually related. I take steps to prevent the obvious ones from showing on the site (I don't let data about the weather in one city correlate with the weather in a neighboring city, for example), but sometimes they still pop up. If they are related, cool! You found a loophole.
    This is exacerbated by the fact that I used "Years" as the base variable. Lots of things happen in a year that are not related to each other! Most studies would use something like "one person" in stead of "one year" to be the "thing" studied.
  3. Observations not independent: For many variables, sequential years are not independent of each other. You will often see trend-lines form. If a population of people is continuously doing something every day, there is no reason to think they would suddenly change how they are doing that thing on January 1. A naive p-value calculation does not take this into account. You will calculate a lower chance of "randomly" achieving the result than represents reality.

    To be more specific: p-value tests are probability values, where you are calculating the probability of achieving a result at least as extreme as you found completely by chance. When calculating a p-value, you need to assert how many "degrees of freedom" your variable has. I count each year (minus one) as a "degree of freedom," but this is misleading for continuous variables.

    This kind of thing can creep up on you pretty easily when using p-values, which is why it's best to take it as "one of many" inputs that help you assess the results of your analysis.
  4. Outliers: Some datasets here have outliers which drag up the correlation. In concept, "outlier" just means "way different than the rest of your dataset." When calculating a correlation like this, they are particularly impactful because a single outlier can substantially increase your correlation.

    Because this page is automatically generated, I don't know whether any of the charts displayed on it have outliers. I'm just a footnote. ¯\_(ツ)_/¯
    I intentionally mishandeled outliers, which makes the correlation look extra strong.



Spurious Scholar was launched January 27, 2024. If you have feedback on it, I'd love to hear from you! Shoot me a note: feedback@tylervigen.com.


Project by Tyler Vigen
emailme@tylervigen.com · about · subscribe


CC BY 4.0