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:

Red-state Republicans and Retail Rendezvous: Correlating Senatorial Support with Searches for Swirling Suds
International Journal of Political Pintonomics
r=0.867 · 95% conf. int. [0.188,0.985] · r2=0.752 · p < 0.05
Generated Jan 2024 · View data details

Got Milk? Exploring the Milky Way: A Holistic Approach to Investigate the Relationship between Milk Consumption and Air Pollution in Canton, Ohio
Journal of Interdisciplinary Dairy Science
r=0.801 · 95% conf. int. [0.628,0.899] · r2=0.642 · p < 0.01
Generated Jan 2024 · View data details

Navigating the Pollution-Recall Nexus: A Steering Study on the Impact of Air Quality in Madison, Indiana
The Journal of Environmental Absurdities
r=0.822 · 95% conf. int. [0.181,0.973] · r2=0.675 · p < 0.05
Generated Jan 2024 · View data details

Breathless in Gulfport: The Air-Quality-Baby Conundrum
Journal of Atmospheric Babysitting and Environmental Quandaries
r=0.924 · 95% conf. int. [0.814,0.970] · r2=0.853 · p < 0.01
Generated Jan 2024 · View data details

The Connection Between Air Pollution and Brickmason Evolution: A Look at Springfield Air From Nine to Five
The Journal of Urban Ecological Engineering
r=0.825 · 95% conf. int. [0.603,0.929] · r2=0.681 · p < 0.01
Generated Jan 2024 · View data details

Clear Skies, Full Theaters: Investigating the Relationship Between Air Quality and Box Office Success in Blacksburg, Virginia
Journal of Atmospheric Aesthetics
r=0.848 · 95% conf. int. [0.421,0.967] · r2=0.719 · p < 0.01
Generated Jan 2024 · View data details

Air-pollution and Money in 'Charleston's Web': A Nitty-Gritty Analysis of Lloyds Banking Group's Stock Price and the Ambient Air Quality in Charleston, West Virginia
Journal of Ecological Economics and Finance
r=0.802 · 95% conf. int. [0.575,0.915] · r2=0.644 · p < 0.01
Generated Jan 2024 · View data details

Making a Link Between Republican Votes in Alabama and Nathan's Hot Dog Eating Contest: A Wiener Takes All Approach
The Journal of Political Gastronomy
r=0.964 · 95% conf. int. [0.864,0.991] · r2=0.930 · p < 0.01
Generated Jan 2024 · View data details

Net Gains: An Examination of the Relationship Between Republican Presidential Candidate Votes in Rhode Island and Frank Lampard’s Premier League Goal Tally
Journal of Sports Analytics and Political Science
r=0.955 · 95% conf. int. [0.636,0.995] · r2=0.912 · p < 0.01
Generated Jan 2024 · View data details

Navigating Political Tides: A Libertarian's Ship to Shipwrecks
The Journal of Political Maritime Studies
r=0.954 · 95% conf. int. [0.811,0.989] · r2=0.910 · p < 0.01
Generated Jan 2024 · View data details

Grate Expectations: The Curious Case of American Cheese Consumption and Republican Votes in Iowa
The Journal of Dairy Politics and Culinary Sociology
r=0.947 · 95% conf. int. [0.730,0.991] · r2=0.897 · p < 0.01
Generated Jan 2024 · View data details

Pawsitively Political: A Feline-Fueled Analysis of Google Searches for 'Cat Memes' and Democrat Votes for Senators in California
The Quarterly Journal of Feline Behavior and Political Science
r=0.989 · 95% conf. int. [0.901,0.999] · r2=0.979 · p < 0.01
Generated Jan 2024 · View data details

The Airbag Anomaly: A Libertarian Link to Recalls Revealed
The Journal of Unconventional Safety Studies
r=0.935 · 95% conf. int. [0.675,0.988] · r2=0.874 · p < 0.01
Generated Jan 2024 · View data details

Shaky Senators: The Earthquake-Electoral Connection in Delaware
Journal of Political Seismology
r=0.947 · 95% conf. int. [0.587,0.994] · r2=0.897 · p < 0.01
Generated Jan 2024 · View data details

Deep Looking at Halloween: A Costumed Connection between YouTube Video Titles and Wisconsin's Costume Attendances
The Journal of Spooky Sociology
r=0.939 · 95% conf. int. [0.634,0.991] · r2=0.881 · p < 0.01
Generated Jan 2024 · View data details

The Existential Comic Conundrum: An Analysis of xkcd Comics and Its Relationship with Average Comment Counts on MinuteEarth YouTube Videos
The Journal of Internet Humor Studies
r=0.864 · 95% conf. int. [0.550,0.964] · r2=0.747 · p < 0.01
Generated Jan 2024 · View data details

Gender Gap in Game Theory: Grasping the Grandiosity of Gender Studies Graduates' Gain on The Game Theorists
The Journal of Playful Economics
r=0.966 · 95% conf. int. [0.869,0.991] · r2=0.932 · p < 0.01
Generated Jan 2024 · View data details

Sew Likeable: Exploring the Correlation Between Sewing Machine Operators in Iowa and Computerphile YouTube Video Likes
The Journal of Domestic Arts and Digital Engagement
r=0.898 · 95% conf. int. [0.618,0.976] · r2=0.807 · p < 0.01
Generated Jan 2024 · View data details

Smart Video Titles and Power People Fights: A Correlation Analysis Delights
The Journal of Whimsical Psychometrics
r=0.902 · 95% conf. int. [0.593,0.979] · r2=0.813 · p < 0.01
Generated Jan 2024 · View data details

YouTube Views and Gangnam Style Cues: A Rhyme in Time
The International Journal of Music and Internet Culture
r=0.992 · 95% conf. int. [0.971,0.998] · r2=0.984 · p < 0.01
Generated Jan 2024 · View data details

Fueling the Vote: An Unlikely Alliance Between Electoral Preferences and Petroleum Pumping Patterns
International Journal of Petroleum Politics and Public Opinion
r=0.976 · 95% conf. int. [0.906,0.994] · r2=0.952 · p < 0.01
Generated Jan 2024 · View data details

Planet Politics: The Pluto Paradox - A Correlational Study of Republican Senatorial Votes in Montana and Google Searches for 'Is Pluto a Planet?'
The Journal of Interplanetary Governance
r=0.910 · 95% conf. int. [0.377,0.990] · r2=0.828 · p < 0.05
Generated Jan 2024 · View data details

Google Searches for 'Please Clap' and Libertarian Votes: A Statistical Rap on Mississippi's Political Map
The Journal of Quirky Political Analyses
r=0.889 · 95% conf. int. [0.032,0.993] · r2=0.791 · p < 0.05
Generated Jan 2024 · View data details

Clearing the Air: Unraveling the Relationship Between Air Pollution in Los Alamos and Kerosene Consumption in Belize
Journal of Environmental Geography
r=0.970 · 95% conf. int. [0.939,0.986] · r2=0.942 · p < 0.01
Generated Jan 2024 · View data details

The Ted-dy Bear Effect: The Unbearable Lightness of Ted in Vernal, Utah
The Journal of Whimsical Anthropology
r=0.803 · 95% conf. int. [0.475,0.935] · r2=0.645 · 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