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:

Dirty Laundry and Right-Wing Sway: The Grand Old Party's Clean Sweep in Missouri
The Journal of Political Paradoxes
r=0.993 · 95% conf. int. [0.935,0.999] · r2=0.986 · p < 0.01
Generated Jan 2024 · View data details

The Mapless Craze: Does the 'Maps Without New Zealand' Meme Influence the Men in Green? A Texas Case Study
The Cartographic Quandary Journal
r=0.854 · 95% conf. int. [0.633,0.946] · r2=0.729 · p < 0.01
Generated Jan 2024 · View data details

Jet Fuel, Physics Feuds, and Nerdy Moods: Analyzing the Relationship Between minutephysics YouTube Video Titles and Jet Fuel Consumption in Nigeria
The Journal of Quirky Science Studies
r=0.848 · 95% conf. int. [0.504,0.960] · r2=0.719 · p < 0.01
Generated Jan 2024 · View data details

Air No Laughing Matter: Uncovering the Unexpected Link between Air Pollution in Wausau and Jet Fuel Usage in Niue
Journal of Ecological Connections
r=0.963 · 95% conf. int. [0.805,0.994] · r2=0.928 · p < 0.01
Generated Jan 2024 · View data details

Driving the Success: Exploring the 'Success Kid' Meme Popularity and Its Impact on City Bus Driver Employment in New Mexico
Journal of Internet Culture and Social Phenomena
r=0.873 · 95% conf. int. [0.677,0.954] · r2=0.763 · p < 0.01
Generated Jan 2024 · View data details

Breath of Fresh Houghton: The Air-y Connection Between Air Quality in Houghton, Michigan, and the Insightfulness of MinuteEarth YouTube Video Titles
The Journal of Ecological Quirkiness
r=0.935 · 95% conf. int. [0.743,0.985] · r2=0.874 · p < 0.01
Generated Jan 2024 · View data details

Air Knowledge: Unraveling the Ties between Air Pollution in Redding, CA and Jet Fuel in Saint Vincent/Grenadines
The Journal of Atmospheric Oddities
r=0.878 · 95% conf. int. [0.735,0.946] · r2=0.771 · p < 0.01
Generated Jan 2024 · View data details

Jenny and the Air Pollution Connection: The Hazy History of Hartford, Connecticut
The Journal of Environmental Anecdotes
r=0.902 · 95% conf. int. [0.826,0.946] · r2=0.814 · p < 0.01
Generated Jan 2024 · View data details

Meme Magic and Maltese Gas: Exploring the Link between 'Pepe' Popularity and Propane Profusion
Journal of Internet Culture and Atmospheric Chemistry
r=0.953 · 95% conf. int. [0.867,0.984] · r2=0.908 · p < 0.01
Generated Jan 2024 · View data details

Fueling Views: The Gas-tastic Connection Between Average Views of Extra History YouTube Videos and Petroleum Consumption in Belize
The Journal of Zany Energy Studies
r=0.952 · 95% conf. int. [0.804,0.989] · r2=0.906 · p < 0.01
Generated Jan 2024 · View data details

Shocking Developments: The Electrifying Link Between Air Quality in Sevierville, Tennessee and Automotive Recalls for Issues with the Electrical System
The Journal of Atmospheric Funktology
r=0.815 · 95% conf. int. [0.657,0.904] · r2=0.663 · p < 0.01
Generated Jan 2024 · View data details

From Math to Meteorology: Connecting the Dots Between 3Blue1Brown YouTube Video Titles and Drenching Rain in New York
The Journal of Interdisciplinary Mathematics and Atmospheric Science
r=0.804 · 95% conf. int. [0.229,0.963] · r2=0.647 · p < 0.05
Generated Jan 2024 · View data details

Soybean GMOs: A Sow of YouTube's Lengthy Crow Show
The Journal of Agricultural Genetics and Social Media Studies
r=0.919 · 95% conf. int. [0.687,0.981] · r2=0.845 · p < 0.01
Generated Jan 2024 · View data details

Kerosene Connections: Illuminating the Relationship Between Libyan Fuel Use and YouTube Engagement
The International Journal of Energy Sociology
r=0.866 · 95% conf. int. [0.555,0.965] · r2=0.750 · p < 0.01
Generated Jan 2024 · View data details

The Shrek Search: Senatorial Selections and Swampy Screen Skims in Silver State Sunshine
The Journal of Quirky Political Science
r=0.895 · 95% conf. int. [0.306,0.989] · r2=0.801 · p < 0.05
Generated Jan 2024 · View data details

Thaddeus and the Pursuit of Libertarian Votes: A Name-candidate Connection Analysis
Journal of Humorous Political Psychology
r=0.910 · 95% conf. int. [0.683,0.977] · r2=0.828 · p < 0.01
Generated Jan 2024 · View data details

Democrat Votes in Florida: Do They Relate to the State's Septic State?
Journal of Political Hygiene
r=0.925 · 95% conf. int. [0.457,0.992] · r2=0.856 · p < 0.01
Generated Jan 2024 · View data details

Lighting Up the Political Landscape: An Illuminating Connection Between Votes for the Republican Presidential Candidate in Illinois and Kerosene Consumption in Eritrea
The Journal of Eclectic Political Analysis
r=0.925 · 95% conf. int. [0.565,0.989] · r2=0.855 · p < 0.01
Generated Jan 2024 · View data details

The Wheezy Connection: A Breath of Fresh Air on US Household Spending and Air Pollution in St. Louis
The Journal of Ecological Economics and Environmental Policy
r=0.859 · 95% conf. int. [0.692,0.939] · r2=0.738 · p < 0.01
Generated Jan 2024 · View data details

Blending Fun and Employment: The Incredibly Whirled Connection Between Simone Giertz's YouTube Video Titles and Blender Tenders in Montana
The Journal of Quirky Innovation and Occupational Integration
r=0.957 · 95% conf. int. [0.651,0.995] · r2=0.916 · p < 0.01
Generated Jan 2024 · View data details

Air Pollution and Duration: A Rhyme-Worthy Connection between Bishop, California and MinuteEarth YouTube Vids
The Journal of Eclectic Environmental Entanglements
r=0.918 · 95% conf. int. [0.707,0.979] · r2=0.842 · p < 0.01
Generated Jan 2024 · View data details

Channeling Energy from the Cosmos: The Provocative Power of SciShow Space YouTube Video Titles in Predicting Geothermal Energy Generation in Italy
The Cosmic Curiosity Journal
r=0.912 · 95% conf. int. [0.579,0.984] · r2=0.831 · p < 0.01
Generated Jan 2024 · View data details

Total Views on Trippy Numberphile and The Tricky Tally of Special Education Teachers in Alabama: A Ten-Year Tale
The Journal of Abstract Arithmetics
r=0.987 · 95% conf. int. [0.949,0.997] · r2=0.974 · p < 0.01
Generated Jan 2024 · View data details

Steve's Screen Time and Estonia's Sunbeam: A Rhyme Time Connection Between YouTube Views and Solar Power
Journal of Rhyming Research
r=0.975 · 95% conf. int. [0.916,0.993] · r2=0.951 · p < 0.01
Generated Jan 2024 · View data details

Gas Emission Association: Bay City's Air Quality and Moldova's LPG Utility
Journal of Atmospheric Chemistry and Urban Environmental Studies
r=0.902 · 95% conf. int. [0.795,0.955] · r2=0.814 · 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