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 Inflated Effects of Air Pollution: A Breath of Fresh Air from Bismarck, North Dakota
Journal of Environmental Pseudoscience
r=0.771 · 95% conf. int. [0.572,0.884] · r2=0.594 · p < 0.01
Generated Jan 2024 · View data details

In Pursuit of Pollution: The Parallels of Air Quality in Albuquerque and Kerosene Usage in Bolivia
The International Journal of Ecological Connections
r=0.733 · 95% conf. int. [0.552,0.848] · r2=0.538 · p < 0.01
Generated Jan 2024 · View data details

I Can't Even Believe It: The Correlation between GMO Cotton in Arkansas and Google Searches for 'I Can't Even'
The Journal of Agricultural Memetics
r=0.867 · 95% conf. int. [0.682,0.948] · r2=0.752 · p < 0.01
Generated Jan 2024 · View data details

Cottoning On: The GMO Connection Between Cotton in Georgia and 'I Have the Flu' Google Searches
The Journal of Ecological Connections
r=0.841 · 95% conf. int. [0.626,0.937] · r2=0.708 · p < 0.01
Generated Jan 2024 · View data details

Blown Away: Unveiling the Celestial Influence on Wind Power in the United Kingdom
The Journal of Atmospheric Jocularity
r=0.940 · 95% conf. int. [0.882,0.970] · r2=0.883 · p < 0.01
Generated Jan 2024 · View data details

Eclipsing Expectations: The Enlightening Connection Between Ezequiel's Popularity and Nepal's Solar Power
The Journal of Sociological Surprises
r=0.976 · 95% conf. int. [0.926,0.992] · r2=0.952 · p < 0.01
Generated Jan 2024 · View data details

Blowin' in the Wind: An Unconventional Connection Between Wind Power in Taiwan and Single Fathers in the United States
The Journal of Eclectic Renewable Energy Research
r=0.975 · 95% conf. int. [0.941,0.990] · r2=0.952 · p < 0.01
Generated Jan 2024 · View data details

Got Milk...and Power? Exploring the Correlation Between Annual US Household Spending on Dairy Products and Hydropower Energy Generated in Nicaragua
Journal of Dairy Economics and Energy Sustainability
r=0.660 · 95% conf. int. [0.331,0.846] · r2=0.436 · p < 0.01
Generated Jan 2024 · View data details

Powering Through: The Sizzling Link Between Hydopower Generation in Estonia and the Consumption of Nathan's Hot Dog Eating Competition Champions
The Journal of Hydrodynamic Gastronomy
r=0.854 · 95% conf. int. [0.713,0.929] · r2=0.729 · p < 0.01
Generated Jan 2024 · View data details

Giddy Geothermal Griddle: An Analysis of the Association between Austria's Geothermal Power and Arson in the United States
Journal of Quirky Geoscience and Social Dynamics
r=0.966 · 95% conf. int. [0.905,0.988] · r2=0.933 · p < 0.01
Generated Jan 2024 · View data details

Illuminating the Connection: Shedding Light on the Link Between Solar Power Generation in Ecuador and Google Searches for 'smol'
Journal of Solar-Powered Internet Trends
r=0.955 · 95% conf. int. [0.877,0.984] · r2=0.912 · p < 0.01
Generated Jan 2024 · View data details

Peculiar Pollution: Paducah, Kentucky and the Perplexing Performance of Bristol-Myers Squibb's Stock Price
The Journal of Eccentric Environmental Economics
r=0.967 · 95% conf. int. [0.874,0.992] · r2=0.935 · p < 0.01
Generated Jan 2024 · View data details

The Pitt and the Pollution: Exploring the Relationship Between Air Quality in Central City, Kentucky and Brad Pitt's Net Worth
The Journal of Celebrity Economics and Environmental Health
r=0.884 · 95% conf. int. [0.678,0.961] · r2=0.781 · p < 0.01
Generated Jan 2024 · View data details

The Hazy Link Between Air Pollution and Criminal Tendencies: A Case Study in St. Marys, Pennsylvania
Journal of Quirky Environmental Psychology
r=0.853 · 95% conf. int. [0.720,0.925] · r2=0.727 · p < 0.01
Generated Jan 2024 · View data details

Smog and the Silver Screen: A Statistical Analysis of Air Pollution in Seneca, South Carolina and Box Office Performance of the Top Movie of the Year
The Journal of Environmental Entertainment Research
r=0.713 · 95% conf. int. [0.267,0.908] · r2=0.509 · p < 0.01
Generated Jan 2024 · View data details

In the Air Tonight: Unraveling the Link Between Ottawa's Air Pollution and Former Czechoslovakia's Jet Fuel Consumption
The Journal of Transcontinental Atmospheric Science
r=0.897 · 95% conf. int. [0.684,0.969] · r2=0.805 · p < 0.01
Generated Jan 2024 · View data details

S'mores Search and Air Pollution in Bozeman: A Beacon on the Correlation
Journal of Gastronomical Ecology
r=0.692 · 95% conf. int. [0.255,0.894] · r2=0.479 · p < 0.01
Generated Jan 2024 · View data details

Jamarcus Jackpots and Banco Bilbao Vizcaya Argentaria's Bucks: Exploring the Connection
The Journal of Financial Linguistics
r=0.917 · 95% conf. int. [0.804,0.966] · r2=0.842 · p < 0.01
Generated Jan 2024 · View data details

Blair-y Interesting: The Surprising Relationship Between the Popularity of the First Name Blair and O'Reilly Automotive's Stock Price (ORLY)
The Journal of Quirky Socioeconomic Trends
r=0.975 · 95% conf. int. [0.939,0.990] · r2=0.952 · p < 0.01
Generated Jan 2024 · View data details

URI Up or URI Down: Exploring the Ruth-URI Connection in the Stock Market
The Journal of Finance and Punny Business
r=0.962 · 95% conf. int. [0.906,0.985] · r2=0.925 · p < 0.01
Generated Jan 2024 · View data details

The Stock Price Link to Name Fame: How Sadie's Popularity Affects WMB Game
Journal of Behavioral Finance and Popular Culture
r=0.819 · 95% conf. int. [0.600,0.924] · r2=0.671 · p < 0.01
Generated Jan 2024 · View data details

The Tactical Tie: Tracing the Ties Between Bachelor's Degrees in Military Technologies and Amazon.com's AMZN Stock Price
The Journal of Military Tech Finance and Fashion
r=0.982 · 95% conf. int. [0.925,0.996] · r2=0.965 · p < 0.01
Generated Jan 2024 · View data details

Ore Edu-cation or Ore Speculation? Exploring the Connection Between Associate Degrees in Education and Freeport-McMoRan's Stock Price
The Journal of Comparative Investucation
r=0.848 · 95% conf. int. [0.505,0.960] · r2=0.719 · p < 0.01
Generated Jan 2024 · View data details

Shining a Light on Solar Power: Illuminating the Connection Between Motor Vehicle Thefts in Missouri and Solar Energy in Tunisia
Journal of Eclectic Energy Studies
r=0.985 · 95% conf. int. [0.947,0.996] · r2=0.971 · p < 0.01
Generated Jan 2024 · View data details

The Fiery Connection: Arson in Minnesota and the Post-Al Service
The Journal of Criminal Pyrotechnics
r=0.954 · 95% conf. int. [0.884,0.982] · r2=0.909 · 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