The Language that Gets People to Give: Phrases that Predict Success on Kickstarter Tanushree Mitra & Eric Gilbert What makes some projects succeed while others fail ? Predictive Features of success and failure ? QUANTITATIVE APPROACH Independent Variables Predictive Features Statistical Model Dependent Variable: Project outcome (Funded or Not) QUANTITATIVE APPROACH Independent Variables Predictive Features(?) Statistical Model Dependent Variable: Project outcome (Funded or Not) Category Video Present Goal Duration Facebook Connected Pitch DATA 45,815K Kickstarter projects all projects as of June 2012 51.53% funded 48.47% not funded 45K Kickstarter project URLs [uni,bi,tri]-grams Fetch project end date phrase frequency > 50? Project reached end date? phrase in all 13 categories? Scrape project description Scrape control variables Lowercase text Remove stop words Penalized Logistic Regression 45K Kickstarter project URLs [uni,bi,tri]-grams Fetch project end date phrase frequency > 50? Project reached end date? phrase in all 13 categories? Scrape project pitch Scrape control variables Lowercase text Remove stop words Statistical Model 45K Kickstarter project URLs [uni,bi,tri]-grams Fetch project end date phrase frequency > 50? Project reached end date? phrase in all 13 categories? Scrape project pitch Scrape control variables Lowercase text Remove stop words Statistical Model 45K Kickstarter project URLs [uni,bi,tri]-grams Fetch project end date phrase frequency > 50? Project reached end date? phrase in all 13 categories? Scrape project pitch Scrape control variables Lowercase text Remove stop words Statistical Model 45K Kickstarter project URLs [uni,bi,tri]-grams Fetch project end date phrase frequency > 50? Project reached end date? phrase in all 13 categories? Scrape project pitch Scrape control variables Lowercase text Remove stop words Statistical Model 45K Kickstarter project URLs [uni,bi,tri]-grams Fetch project end date phrase frequency > 50? Project reached end date? phrase in all 13 categories? Scrape project pitch Scrape control variables Lowercase text Remove stop words Statistical Model ~20K unigrams, bigrams, trigrams Pitch 45K Kickstarter project URLs [uni,bi,tri]-grams Fetch project end date phrase frequency > 50? Project reached end date? phrase in all 13 categories? Scrape project pitch 59 control variables Scrape control variables Lowercase text Remove stop words Statistical Model Category Video Present Goal Duration Facebook Connected 45K Kickstarter project URLs [uni,bi,tri]-grams Fetch project end date phrase frequency > 50? Project reached end date? phrase in all 13 categories? Scrape project pitch 59 control variables Scrape control variables Lowercase text Remove stop words Statistical Model STATISTICAL TECHNIQUE Independent Variables Phrases (20K) + Controls (59) Statistical Model Dependent Variable: Project outcome (Funded or Not) STATISTICAL TECHNIQUE Independent Variables Phrases (20K) + Controls (59) Penalized Logistic Regression Dependent Variable: Project outcome (Funded or Not) Friedman et al. 2010 Results: MODEL FITS Baseline Model | | Error 48.47 % Controls Only Model Explanatory Power | Phrases + Controls Model Explanatory Power | 40.8 % 58.56 % | | Error 17.03% Error 2.24% (NF) Predictors not been able ( β = − 4.07 ) “I have not been able to finish the film because none of my editors will see the project through to the end.” (NF) Predictors later i ( β = − 3.04 ) hope to get ( β = − 2.39 ) “I can’t take size orders and possibly hope to get them all made in time for christmas.” (NF) Predictors even a dollar ( β = − 3.10 ) Wattenberg & VieĢgas, 2008 (F) Predictors mention your ( β = 2.69 ) also receive ( β = 1.83 ) add $40 and you will also receive two vip tickets to the premiere screening. (F) Predictors next step is ( β = 1.07 ) Recording is pretty much done, next step is production. (F) Predictors cats ( β = 2.64 ) UNDERSTANDING CONTEXT A closer look at predictive phrases Principles of Persuasion Cialdini, R. B. 1993 Principles of Persuasion 1. 2. 3. 4. 5. 6. Reciprocity Scarcity Authority Social Proof Social Identity Liking Cialdini, R. B. 1993 Principles of Persuasion 1. 2. 3. 4. 5. 6. Reciprocity Scarcity Authority Social Proof Social Identity Liking Cialdini, R. B. 1993 RECIPROCITY Brehm & Cole 1966, Goranson & Berkowitz, 1966, Ciladini 2001 RECIPROCITY we’ll mention your (β = 2.69) name in the sleeve of our full length album RECIPROCITY I will thank you on my website, send you good karma and (β = 2.04) .. SCARCITY Ciladini 2001, Ciladini & Goldstein 2004 SCARCITY also, you will be given the chance (β = 2.69) to purchase our small batch pieces before the public domain AUTHORITY Ciladini 2001, Ciladini & Goldstein 2004 AUTHORITY the project will be (β = 18.48) produced by dove award winning producer SOCIAL PROOF Ciladini 2001 SOCIAL PROOF [name] has pledged (β = 5.42) some money..… so, you can see that i already have people willing to support my art. Language is a reliable signal of success of crowd-funded projects So are some controls…. CONTROLS CONTROLS Graphic Design β = 1.35 Illustration β = -2.55 Video Present β = 0.60 Journalism β = -1.12 Project Duration β = -0.01 Facebook Connected β = 0.13 … … IMPLICATIONS IMPLICATIONS http://www.cc.gatech.edu/~tmitra3/data/KS.predicts Also at: http://b.gatech.edu/1mf1C6E The Language that Gets People to Give: Phrases that Predict Success on Kickstarter Tanushree Mitra & Eric Gilbert @tanmit | @eegilbert DATA: http://b.gatech.edu/1mf1C6E Fancy Stats. Huh! Google 1T Corpus phrases Scan presence of KS phrases in Google 1T χ2 test between phrase frequencies + Bonferroni Correction Search for phrases with significantly higher difference + Membership higher in Ggle 1T 54K Kickstarter phrases Non-zero β weights GENERAL PHRASES: - 494 positive predictors - 453 negative predictors next step is in the upcoming to announce provide us need one ….