Debunking Other People's Data Science

! DEBUNKING OTHER PEOPLE’S DATA SCIENCE

@bellmar 2 Marianne Bellotti I build data infrastructure Senior Software
Developer The United Nations : Humanitarian Data Exchange CTO & Founder Exversion: Version Control for Data 1EECUKQPCN " DCFCUU

@bellmar 3 I lie with data

@bellmar 4

@bellmar 5 “The wonderful thing about being a data scientist
is that I get all of the credibility of genuine science, with none of the irritating peer review or reproducibility worries ... I thought I was publishing an entertaining view of some data I'd extracted, but it was treated like a scientiﬁc study.” Pete Warden

@bellmar 6 Pete Warden’s Books

@bellmar 7

@bellmar 8 Data science is about exploring simple components of
complex things.

@bellmar # We Trust Science, But Do Not Enforce Scientiﬁc
Methods Keynote Template

@bellmar $ 35% of Data Scientists think “data science” is
a buzz word Keynote Template

a buzz word Keynote Template 25+OCFGVJKUWR

a buzz word Keynote Template FQPņV! FQPņV FQPņVMPQY FQPņVJCXGCPQRKPKQP FKFPņVTGURQPF

@bellmar % Reporting only part of the analysis Keynote Template

@bellmar $ What information is missing? What data points weren’t
collected? Keynote Template

@bellmar $ How are things deﬁned? Keynote Template

a buzz word Keynote Template

Se7en - Creative Keynote Template 17 Who are data scientists?
Counting is harder than you think Population Stats Hackers Others Statisticians Analysts Data Tools Data Warehousing Academics Journalists Data Scientists DATA! &

@bellmar ? Silicon Valley -vs- Silicon Alley Keynote Template

@bellmar 19

@bellmar 20 >>> import pandas as pd >>> import numpy
as np >>> import matplotlib as plot

@bellmar 21 Free Text is the Enemy of Data

@bellmar 22 Where is Silicon Valley?

@bellmar 23

@bellmar 24 What about San Francisco? '

@bellmar 25 “a third of Silicon Valley scientists and engineers
are immigrants” ! ! “females outnumber males in the majority of the localities in Silicon Valley, and the proportion of women never drops under 49%, with the exception of Milpitas, with 48.9%” ! ! “Silicon Valley has the most millionaires and the most billionaires in the United States per capita.”

@bellmar ! Misleading deﬁnitions Keynote Template

@bellmar 27

@bellmar 28 O RLY?

@bellmar 29 Important to note, the BDS (Business Dynamics Statistics)
tracks employer-ﬁrms and excludes the self-employed or non-employer ﬁrms. For more on the BDS, see U.S. Census Bureau Center for Economic Studies

@bellmar 30 Entrepreneurship is Dying! ! -vs- ! Increase in
sole proprietor, freelance situations

@bellmar 31

@bellmar ! Failing Fifth Grade Math Keynote Template

@bellmar 33

@bellmar 34

@bellmar 35 75% 40%

@bellmar | How to Avoid Criticism Keynote Template

@bellmar # Clearly deﬁne your assumptions Keynote Template

@bellmar 38 In 2003, the UK government estimated that there
were 4,000 traﬃcked women in the country

@bellmar 39 The actual number of traﬃcked women from police
reports was 71

@bellmar 40 The original estimate had been calculated by counting
the TOTAL number of female non-UK citizens working in the sex industry and assuming they were ALL victims of traﬃcking

@bellmar 41 Want To Know The Language Of The Future?
The Data Suggests It Could Be...French

@bellmar # Make process part of the discussion, not the
footnotes Keynote Template

@bellmar 43 Data doesn’t tell stories, people do

@bellmar # Aim for replication Keynote Template

@bellmar # Test for basic skews, sampling errors and other
factors Keynote Template

@bellmar 46

@bellmar 47 Junky: Dataset Validation https://github.com/exversion/junky

@bellmar # Knowing how your data was collected and explaining
that to people is part of the analysis Keynote Template

@bellmar 49 Data scientists like to play with data! It’s
okay to lie, just let everyone else in on it

$ THANK YOU!

Debunking Other People's Data Science

Debunking Other People's Data Science

More Decks by Marianne Bellotti

Other Decks in Science

Featured

Transcript