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:

Jace's Judicious Juxtaposition: Unraveling the UFO Connection in Idaho
The Journal of Extraterrestrial Enigmas and Regional Anomalies
r=0.934 · 95% conf. int. [0.883,0.963] · r2=0.872 · p < 0.01
Generated Jan 2024 · View data details

UFO a Summit: A Statistical Analysis of the Relationship between UFO Sightings in Alabama and Total Number of Successful Mount Everest Climbs
The Journal of Extraterrestrial Studies and Mountain Metrics
r=0.892 · 95% conf. int. [0.798,0.943] · r2=0.795 · p < 0.01
Generated Jan 2024 · View data details

Out of this World Correlations: Unveiling the Interstellar Connection Between UFO Sightings in Missouri and Petroleum Consumption in Kuwait
The International Journal of Extraterrestrial Science and Geopolitical Relations
r=0.902 · 95% conf. int. [0.824,0.946] · r2=0.813 · p < 0.01
Generated Jan 2024 · View data details

Heli-calling for Help: Analyzing the Relationship Between Telephone Operators in Oklahoma and Google Searches for 'Helicopter Accident'
The Journal of Quirky Research Studies
r=0.896 · 95% conf. int. [0.721,0.964] · r2=0.803 · p < 0.01
Generated Jan 2024 · View data details

Associates in Architecture: Analyzing the Affect on Pediatrician Population in New Mexico
The Journal of Architectural Anthropology
r=0.826 · 95% conf. int. [0.447,0.953] · r2=0.682 · p < 0.01
Generated Jan 2024 · View data details

Dead Ringers: The Correlation Between Athletes and Google Searches for 'Zombies' in Florida
The Journal of Mysterious Sports Trends
r=0.871 · 95% conf. int. [0.689,0.949] · r2=0.758 · p < 0.01
Generated Jan 2024 · View data details

The Trade of Trimmers and Troubles at Sea: A Tale of Two Tides
The Journal of Nautical Nonsense and Maritime Mischief
r=0.802 · 95% conf. int. [0.295,0.957] · r2=0.643 · p < 0.01
Generated Jan 2024 · View data details

Got Milk? Examining the Udderly Bizarre Link Between Milk Consumption and Burglaries in Delaware
Journal of Dairy Delinquency Studies
r=0.869 · 95% conf. int. [0.747,0.935] · r2=0.756 · p < 0.01
Generated Jan 2024 · View data details

Robberies in Kansas and Remaining Forest Canvases: A Surprising Link Revealed
The Journal of Ecological Criminology
r=0.887 · 95% conf. int. [0.789,0.941] · r2=0.787 · p < 0.01
Generated Jan 2024 · View data details

Genetically Modified Soybeans in Kansas and the Gluttonous Gluttony: Gobbling GMOs and Hotdog Hurling
The Journal of Agricultural Absurdities
r=0.853 · 95% conf. int. [0.681,0.936] · r2=0.728 · p < 0.01
Generated Jan 2024 · View data details

Maize Mania: The Corny Connection Between GMOs and Compensation Managers in Michigan
The Journal of Agri-Business and Human Resources Compensation
r=0.937 · 95% conf. int. [0.839,0.976] · r2=0.877 · p < 0.01
Generated Jan 2024 · View data details

Burnin' Love: A Fiery Connection Between Arson in Arizona and Kerosene Consumption in Turkiye
Journal of Pyrostatistics and Combustion Dynamics
r=0.850 · 95% conf. int. [0.728,0.920] · r2=0.722 · p < 0.01
Generated Jan 2024 · View data details

The Correlation Between the Spacing of Saturn and Uranus and the Surprising Surge of UFO Sightings in California: A Celestial Connection
Journal of Extraterrestrial Phenomena Research
r=0.902 · 95% conf. int. [0.830,0.945] · r2=0.814 · p < 0.01
Generated Jan 2024 · View data details

UFO or Fossil Fuels: Unveiling the Interstellar Link
The Journal of Extraterrestrial Energy Studies
r=0.847 · 95% conf. int. [0.731,0.915] · r2=0.717 · p < 0.01
Generated Jan 2024 · View data details

Cartographers in Oklahoma and xkcd Comics: A Rhyme in Space and Time
The Journal of Geographical Humor and Pop Culture
r=0.677 · 95% conf. int. [0.273,0.878] · r2=0.458 · p < 0.01
Generated Jan 2024 · View data details

Spreading Butter, Building Bridges: A Creamy Correlation Between Butter Consumption and the Number of Bridge and Lock Tenders in Texas
The Journal of Culinary Correlations
r=0.736 · 95% conf. int. [0.423,0.892] · r2=0.541 · p < 0.01
Generated Jan 2024 · View data details

Zaria-n Landscapes: Exploring the Correlation between the Popularity of the Name Zaria and the Number of Landscape Architects in Connecticut
Journal of Quirky Amalgamations
r=0.707 · 95% conf. int. [0.386,0.876] · r2=0.501 · p < 0.01
Generated Jan 2024 · View data details

The Aliza Effect: Exploring the Curious Correlation Between the Popularity of the Name Aliza and the Number of Statisticians in Maine
The Journal of Quirky Statistics and Unlikely Correlations
r=0.780 · 95% conf. int. [0.515,0.909] · r2=0.608 · p < 0.01
Generated Jan 2024 · View data details

Ahmed: A Name to Engineer By? Investigating the Correlation Between Ahmed's Popularity and University Engineering Teachers in Georgia
The Journal of Cultural Naming Studies
r=0.888 · 95% conf. int. [0.712,0.959] · r2=0.789 · p < 0.01
Generated Jan 2024 · View data details

Shoe-niqua and the Cobbler Connection: An Empirical Study of the Relationship between the Popularity of the Name Shaniqua and the Number of Shoe and Leather Workers and Repairers in Maine
Journal of Sociolinguistics and Occupational Demographics
r=0.855 · 95% conf. int. [0.379,0.973] · r2=0.731 · p < 0.01
Generated Jan 2024 · View data details

Ludicrous Link: Larcenies in Montana and Linearity with Days of Our Lives' Loyalists
The Journal of Eccentric Behavioral Science
r=0.932 · 95% conf. int. [0.872,0.965] · r2=0.869 · p < 0.01
Generated Jan 2024 · View data details

When UFOs Light up the Californian Sky, Vegas Hotels Reach for the Stars: An Unexpected Correlation
The Journal of Extraterrestrial Economics and Interstellar Tourism
r=0.925 · 95% conf. int. [0.860,0.960] · r2=0.855 · p < 0.01
Generated Jan 2024 · View data details

Genetically Muddled Offspring: Exploring the Asthma-GMO Cotton Connection in American Children
Journal of Genetic Medicine and Environmental Health
r=0.837 · 95% conf. int. [0.626,0.934] · r2=0.700 · p < 0.01
Generated Jan 2024 · View data details

Grainy Relationship: Wheat Feed Volume and Titanic Search Behavior in the United States
The Journal of Agricultural Psychology and Maritime Sociology
r=0.783 · 95% conf. int. [0.431,0.928] · r2=0.612 · p < 0.01
Generated Jan 2024 · View data details

Maize and Mistranslations: Exploring the Unlikely Relationship Between GMO Corn Cultivation in Indiana and the Surge in Interpretation and Translation Services
The Journal of Transgenic Agriculture and Linguistic Anthropology
r=0.960 · 95% conf. int. [0.899,0.984] · r2=0.921 · 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