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:

Actuarial Attractions: Exploring the Correlation Between LEMMiNO YouTube Video Views and the Number of Number Crunchers in Kansas
The Journal of Quirky Quantitative Studies
r=0.915 · 95% conf. int. [0.697,0.978] · r2=0.836 · p < 0.01
Generated Jan 2024 · View data details

Clearing the Air: Analyzing the Relationship between Air Pollution in Boston and Kerosene Consumption in the United States
The Journal of Environmental Quirks and Quandaries
r=0.865 · 95% conf. int. [0.763,0.925] · r2=0.748 · p < 0.01
Generated Jan 2024 · View data details

The Air-ious Connection: Examining the Correlation between Air Pollution in Atlanta and the Number of Library Assistants in Georgia
The Journal of Ecological Anomalies
r=0.898 · 95% conf. int. [0.757,0.959] · r2=0.807 · p < 0.01
Generated Jan 2024 · View data details

From Kerosene to Kreative Kerosene: Exploring the Correlation between Vihart YouTube Video Titles and Kerosene Usage in Tanzania
The International Journal of Interdisciplinary Kerosene Studies
r=0.684 · 95% conf. int. [0.213,0.897] · r2=0.468 · p < 0.01
Generated Jan 2024 · View data details

Can Tree Nuts on the Go Trigger SmarterYouTube Flow?: Exploring the Relationship Between US Tree Nut Consumption and Total Comments on SmarterEveryDay Videos
The Journal of Nutty Science
r=0.905 · 95% conf. int. [0.732,0.968] · r2=0.819 · p < 0.01
Generated Jan 2024 · View data details

Revealing the Relationship Between Republican Votes for Senators in Virginia and the Popularity of the 'Like a Boss' Meme
Journal of Political Psychology and Internet Culture
r=0.878 · 95% conf. int. [0.230,0.987] · r2=0.770 · p < 0.05
Generated Jan 2024 · View data details

An Ode to Roads and Votes: The Antidote for BMW and Republican Boat Floats
The Journal of Transport Politics and Cultural Studies
r=0.890 · 95% conf. int. [0.645,0.969] · r2=0.791 · p < 0.01
Generated Jan 2024 · View data details

Flexing on Flexibility: The Hip and With It Connection Between AsapSCIENCE YouTube Video Titles and the Demand for Physical Therapist Aides in New York
The Journal of Interdisciplinary YouTube Studies
r=0.864 · 95% conf. int. [0.547,0.964] · r2=0.746 · p < 0.01
Generated Jan 2024 · View data details

Fueling the Fire: The Correlation Between MinuteEarth Video Titles and Gasoline Consumption in Madagascar
The Journal of Comical Ecological Studies
r=0.907 · 95% conf. int. [0.612,0.981] · r2=0.823 · p < 0.01
Generated Jan 2024 · View data details

Burning Bright: Air Quality in Houghton, Michigan and Liquefied Petroleum Gas in Central African Republic - A Gas-tly Connection
Journal of Eclectic Atmospheric Studies
r=0.984 · 95% conf. int. [0.950,0.995] · r2=0.969 · p < 0.01
Generated Jan 2024 · View data details

Airing Out the Sun: Exploring the Affair Between Peoria's Air and Libya's Light
The Journal of Atmospheric Affinities
r=0.904 · 95% conf. int. [0.685,0.973] · r2=0.816 · p < 0.01
Generated Jan 2024 · View data details

Weight in Gold: Unearthing the Correlation Between the Price of Gold and the Total Length of Vihart YouTube Videos
Journal of Improbable Mathematics
r=0.925 · 95% conf. int. [0.454,0.992] · r2=0.855 · p < 0.01
Generated Jan 2024 · View data details

Studying the Asphalt-tively Impactful Relationship Between Transportation Bachelor's Degrees and Air Pollution in Laramie, Wyoming
The Journal of Ecological Engineering and Urban Transportation
r=0.861 · 95% conf. int. [0.504,0.967] · r2=0.741 · p < 0.01
Generated Jan 2024 · View data details

Callie-doscope: Examining the Correlation Between Callie's Popularity and Air Pollution in Mayfield, Kentucky
Journal of Ecological Quirks
r=0.817 · 95% conf. int. [0.484,0.943] · r2=0.667 · p < 0.01
Generated Jan 2024 · View data details

The Paxton Paradox: A Statistical Connection between First Name Popularity and Libertarian Votes for Senators in Wisconsin
The Journal of Quirky Quantitative Research
r=0.968 · 95% conf. int. [0.849,0.993] · r2=0.936 · p < 0.01
Generated Jan 2024 · View data details

A Breath of Fresh Search: The Nose Knows in Evansville, Indiana
The Journal of Experimental Olfactory Science
r=0.819 · 95% conf. int. [0.591,0.926] · r2=0.671 · p < 0.01
Generated Jan 2024 · View data details

Degrees of Precision: A Closer Look at Air Pollution in San Juan, Puerto Rico
Journal of Atmospheric Accuracy
r=0.867 · 95% conf. int. [0.556,0.965] · r2=0.751 · p < 0.01
Generated Jan 2024 · View data details

Connecting Cat Memes and Carbon Monoxide: An Amusing Analysis of Air Pollution in Gainesville, Florida
The Journal of Cat-astrophic Environmental Studies
r=0.901 · 95% conf. int. [0.590,0.979] · r2=0.812 · p < 0.01
Generated Jan 2024 · View data details

A Tale of Books and Smoke: Spoke on US Folks and Tokes in Rocky Mount, NC
The Journal of Appalachian Cultural Studies
r=0.858 · 95% conf. int. [0.602,0.954] · r2=0.737 · p < 0.01
Generated Jan 2024 · View data details

Pump Up the Votes: The Political Correlation Between Democrat Support for Senators in Texas and the Wellhead Pumpers Population
The Journal of Political Demographics and Societal Trends
r=0.971 · 95% conf. int. [0.754,0.997] · r2=0.943 · p < 0.01
Generated Jan 2024 · View data details

Shedding Light on Electoral Illuminations: The Illuminating Connection Between Republican Votes in Minnesota and Automotive Recalls for Exterior Lighting
The Journal of Quirky Interdisciplinary Studies
r=0.912 · 95% conf. int. [0.709,0.975] · r2=0.831 · p < 0.01
Generated Jan 2024 · View data details

Libertarian Lunacy: Linking Presidential Preference to Paraguayan Petroleum Consumption
The Journal of Quirky Quantitative Analysis
r=0.904 · 95% conf. int. [0.600,0.980] · r2=0.817 · p < 0.01
Generated Jan 2024 · View data details

Libertarian Votes in Georgia and Sudan's Fossil Fuel Bonanza: A Rhyming Analysis
The International Journal of Geopolitical Limericks
r=0.866 · 95% conf. int. [0.520,0.968] · r2=0.750 · p < 0.01
Generated Jan 2024 · View data details

Pondering Petroleum: Perusing the Playfulness of Extra History YouTube Video Titles and Petroleum Consumption in Peculiar Greenland
The Journal of Amusing Energy and Unconventional Studies
r=0.904 · 95% conf. int. [0.636,0.977] · r2=0.817 · p < 0.01
Generated Jan 2024 · View data details

Spreading the Love: A Butterly Connection Between Butter Consumption and Average Number of Comments on OverSimplified YouTube Videos
The Journal of Quirky Scientific Studies
r=0.964 · 95% conf. int. [0.703,0.996] · r2=0.930 · 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