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:

Kerosene Consumption in Canada and Libertarian Votes in Alaska: A Correlative Conundrum
The Journal of Quirky Quantitative Analysis
r=0.887 · 95% conf. int. [0.613,0.970] · r2=0.786 · p < 0.01
Generated Jan 2024 · View data details

Wacky Wyoming: Waltzing with Senatorial Selection and Stock Sway
The Journal of Quirky State Politics
r=0.866 · 95% conf. int. [0.184,0.985] · r2=0.750 · p < 0.05
Generated Jan 2024 · View data details

Parking Brake Libertarianism: An Underestimated Relationship Between Political Preferences and Automotive Mishaps
The Journal of Quirky Social Science Research
r=0.969 · 95% conf. int. [0.854,0.994] · r2=0.938 · p < 0.01
Generated Jan 2024 · View data details

The Gaslighting of Libertarian Votes: Uncovering the Link Between Massachusetts Senatorial Preferences and U.S. Virgin Islands Gasoline Consumption
The Journal of Political Paradoxes
r=0.965 · 95% conf. int. [0.709,0.996] · r2=0.932 · p < 0.01
Generated Jan 2024 · View data details

The Cheddar Vote: American Cheese Consumption and Republican Candidate Preference in Michigan
Journal of Dairy Politics
r=0.916 · 95% conf. int. [0.596,0.985] · r2=0.839 · p < 0.01
Generated Jan 2024 · View data details

Ridin' Mississippian Senators: A Statistical Examination of 'That's What She Said' Meme Popularity and Republican Votes
Journal of Memetic Studies
r=0.923 · 95% conf. int. [0.446,0.992] · r2=0.852 · p < 0.01
Generated Jan 2024 · View data details

Corny Politics: Genetically Modified Corn and Libertarian Votes in North Carolina
The Journal of Agrarian Ethics and Biopolitics
r=0.908 · 95% conf. int. [0.488,0.986] · r2=0.824 · p < 0.01
Generated Jan 2024 · View data details

The Air Bags and Libertarians: A Correlation Analysis of Automotive Recalls and Libertarian Presidential Votes in Pennsylvania
The Journal of Quirky Sociology
r=0.959 · 95% conf. int. [0.783,0.993] · r2=0.919 · p < 0.01
Generated Jan 2024 · View data details

Thaddeus for the Libertarianus: An Analysis of the Thaddeus Name Popularity and Presidential Votes in Kansas
Journal of Political Names Research
r=0.941 · 95% conf. int. [0.765,0.986] · r2=0.886 · p < 0.01
Generated Jan 2024 · View data details

Flight of the Elephants: The Trunkated Relationship Between Republican Votes in Illinois and Google Searches for 'Flights to Antarctica'
Journal of Political Pachydermology
r=0.964 · 95% conf. int. [0.698,0.996] · r2=0.929 · p < 0.01
Generated Jan 2024 · View data details

Drilling Down: Unearthing the Connection Between Republican Votes in Oklahoma and Petroleum Consumption in Lebanon
The International Journal of Political Petrology
r=0.851 · 95% conf. int. [0.601,0.950] · r2=0.725 · p < 0.01
Generated Jan 2024 · View data details

Say Cheese: The Gouda, the Bad, and the Ugly - Exploring the Relationship between American Cheese Consumption and Votes for the Republican Presidential Candidate in Indiana
The Cheese Chronicle
r=0.947 · 95% conf. int. [0.727,0.991] · r2=0.896 · p < 0.01
Generated Jan 2024 · View data details

Waste Not, Want Not: A Correlational Examination of AsapSCIENCE Video Titles Trends and Garbage Collector Employment in Mississippi
The Journal of Quirky Social Science Research
r=0.879 · 95% conf. int. [0.591,0.968] · r2=0.773 · p < 0.01
Generated Jan 2024 · View data details

Cleaning Counts: Correlating the Counts of Housekeepers in West Virginia with Comments on SciShow Space
The International Journal of Domestic Hygiene and Astrophysical Anecdotes
r=0.963 · 95% conf. int. [0.830,0.992] · r2=0.928 · p < 0.01
Generated Jan 2024 · View data details

Regina-phobic Revelations: A Statistical Analysis of the Relationship Between the Popularity of the First Name Regina and the Entertainment Value of Casually Explained YouTube Video Titles
The Journal of Lighthearted Linguistics
r=-0.724 · 95% conf. int. [-0.946,-0.040] · r2=0.524 · p < 0.05
Generated Jan 2024 · View data details

iCan't Believe It's Apples and Likes: Exploring the Surprising Connection Between Global Apple iPhone Sales in Q3 and Total Likes of AsapSCIENCE YouTube Videos
The Journal of Digital Fruit Consumption Studies
r=0.893 · 95% conf. int. [0.428,0.984] · r2=0.798 · p < 0.01
Generated Jan 2024 · View data details

The Ties Between Title Insightfulness in Mark Rober's YouTube Videos and the Total Tally of Lawyers in the United States
International Journal of Social Media and Cultural Analysis
r=0.946 · 95% conf. int. [0.815,0.985] · r2=0.895 · p < 0.01
Generated Jan 2024 · View data details

Get with the Times: A Hip Connection Between YouTube Video Titles and 'How to Move to Europe' Google Searches
The Journal of Digital Culture and Media Studies
r=0.813 · 95% conf. int. [0.415,0.950] · r2=0.661 · p < 0.01
Generated Jan 2024 · View data details

Soybean GMOs and Space-Time YouTube Show: The Zeitgeist in Statistical Might
The Journal of Quantum Agriculture and Interstellar Media Studies
r=0.962 · 95% conf. int. [0.797,0.993] · r2=0.925 · p < 0.01
Generated Jan 2024 · View data details

The Ezequiel Effect: Name Popularity and Political Preference in Delaware
The Journal of Sociopolitical Names Studies
r=0.840 · 95% conf. int. [0.602,0.941] · r2=0.705 · p < 0.01
Generated Jan 2024 · View data details

Alice in Voterland: The Curious Case of Libertarian Leanings in Minnesota Senators Named Alice
The Journal of Political Curiosities
r=0.978 · 95% conf. int. [0.897,0.996] · r2=0.957 · p < 0.01
Generated Jan 2024 · View data details

From Red States to Efficient Plates: A Biomass of Political Power in Louisiana and Taiwan
Journal of Cross-Cultural Energy Politics
r=0.987 · 95% conf. int. [0.927,0.998] · r2=0.974 · p < 0.01
Generated Jan 2024 · View data details

Leonel, Lies, and Legislation: The Influence of Name Popularity on Political Preferences in North Carolina
The Journal of Sociopolitical Naming Trends
r=0.911 · 95% conf. int. [0.747,0.970] · r2=0.830 · p < 0.01
Generated Jan 2024 · View data details

Peeling Back the Layers: A Fruitful Investigation into Household Spending on Processed Fruits and Votes for the Democrat Presidential Candidate in Tennessee
Journal of Fruitful Economics
r=0.975 · 95% conf. int. [0.787,0.997] · r2=0.952 · p < 0.01
Generated Jan 2024 · View data details

Wind and Libertarians: A Tandem Too Tempting to Ignore
The Journal of Unusual Political Alliances
r=0.854 · 95% conf. int. [0.573,0.956] · r2=0.730 · 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