Reproducible,Research,with, Evidence5based,DataAnalysis,rpeng/talks/pengslides_transscience.pdf ·...

44
Reproducible Research with Evidencebased Data Analysis Roger D. Peng, PhD Department of Biosta/s/cs Johns Hopkins Bloomberg School of Public Health @rdpeng, @simplystats, simplystatistics.org April 2014

Transcript of Reproducible,Research,with, Evidence5based,DataAnalysis,rpeng/talks/pengslides_transscience.pdf ·...

Page 1: Reproducible,Research,with, Evidence5based,DataAnalysis,rpeng/talks/pengslides_transscience.pdf · Reproducible,Research,with, Evidence5based,DataAnalysis, Roger,D.,Peng,,PhD, Departmentof)Biosta/s/cs)

Reproducible  Research  with  Evidence-­‐based  Data  Analysis  

Roger  D.  Peng,  PhD  Department  of  Biosta/s/cs  

Johns  Hopkins  Bloomberg  School  of  Public  Health  @rdpeng, @simplystats, simplystatistics.org!

!

April  2014  

Page 2: Reproducible,Research,with, Evidence5based,DataAnalysis,rpeng/talks/pengslides_transscience.pdf · Reproducible,Research,with, Evidence5based,DataAnalysis, Roger,D.,Peng,,PhD, Departmentof)Biosta/s/cs)

ReplicaAon,  ReplicaAon  

Replica(on  •  Focuses  on  the  validity  of  the  

scien/fic  claim  •  “Is  this  claim  true?”  •  The  ulAmate  standard  for  

strengthening  scienAfic  evidence  •  New  invesAgators,  data,  

analyAcal  methods,  laboratories,  instruments,  etc.  

•  ParAcularly  important  in  studies  that  can  impact  broad  policy  or  regulatory  decisions  

Page 3: Reproducible,Research,with, Evidence5based,DataAnalysis,rpeng/talks/pengslides_transscience.pdf · Reproducible,Research,with, Evidence5based,DataAnalysis, Roger,D.,Peng,,PhD, Departmentof)Biosta/s/cs)

Underlying  Trends  •  Some  studies  cannot  be  replicated:  No  Ame,  No  money,  Unique/opportunisAc  

•  Technology  is  increasing  data  collecAon  throughput;  data  are  more  complex  and  high-­‐dimensional  

•  ExisAng  databases  can  be  merged  to  become  bigger  databases  (but  data  are  used  off-­‐label)  

•  CompuAng  power  allows  more  sophisAcated  analyses,  even  on  “small”  data  

•  Data  science,  big  data,  you  name  it  •  For  every  field  “X”  there  is  a  “ComputaAonal  X”  

Page 4: Reproducible,Research,with, Evidence5based,DataAnalysis,rpeng/talks/pengslides_transscience.pdf · Reproducible,Research,with, Evidence5based,DataAnalysis, Roger,D.,Peng,,PhD, Departmentof)Biosta/s/cs)

What’s  Wrong  with  ReplicaAon?  

•  Some  studies  cannot  be  replicated  •  No  Ame,  decisions  must  be  made  now  •  No  money  •  OpportunisAc  /  Unique  

Page 5: Reproducible,Research,with, Evidence5based,DataAnalysis,rpeng/talks/pengslides_transscience.pdf · Reproducible,Research,with, Evidence5based,DataAnalysis, Roger,D.,Peng,,PhD, Departmentof)Biosta/s/cs)

Air  PolluAon  and  Health  Research:  A  Perfect  Storm  

•  EsAmaAng  small  health  effects  in  the  presence  of  much  stronger  signals  

•  Results  inform  substanAal  policy  decisions  and  affect  many  stakeholders  

•  EPA  regulaAons  can  cost  billions  of  dollars  

•  Complex  staAsAcal  methods  are  needed  and  subjected  to  intense  scruAny  

Page 6: Reproducible,Research,with, Evidence5based,DataAnalysis,rpeng/talks/pengslides_transscience.pdf · Reproducible,Research,with, Evidence5based,DataAnalysis, Roger,D.,Peng,,PhD, Departmentof)Biosta/s/cs)

The  Result?  

•  Even  basic  analyses  are  difficult  to  describe  •  Heavy  computaAonal  requirements  are  thrust  upon  people  without  adequate  training  in  staAsAcs  and  compuAng  

•  Errors  are  more  easily  introduced  into  long  analysis  pipelines  

•  Knowledge  transfer  is  inhibited  •  Results  are  difficult  to  replicate  or  reproduce  

 Complicated  analyses  cannot  be  trusted  

Page 7: Reproducible,Research,with, Evidence5based,DataAnalysis,rpeng/talks/pengslides_transscience.pdf · Reproducible,Research,with, Evidence5based,DataAnalysis, Roger,D.,Peng,,PhD, Departmentof)Biosta/s/cs)

Recent  Developments  

Page 8: Reproducible,Research,with, Evidence5based,DataAnalysis,rpeng/talks/pengslides_transscience.pdf · Reproducible,Research,with, Evidence5based,DataAnalysis, Roger,D.,Peng,,PhD, Departmentof)Biosta/s/cs)

“DecepAon  at  Duke”  

The  Duke  Saga  

Page 9: Reproducible,Research,with, Evidence5based,DataAnalysis,rpeng/talks/pengslides_transscience.pdf · Reproducible,Research,with, Evidence5based,DataAnalysis, Roger,D.,Peng,,PhD, Departmentof)Biosta/s/cs)

“Rock  Star”  StaAsAcians  

Page 10: Reproducible,Research,with, Evidence5based,DataAnalysis,rpeng/talks/pengslides_transscience.pdf · Reproducible,Research,with, Evidence5based,DataAnalysis, Roger,D.,Peng,,PhD, Departmentof)Biosta/s/cs)

InsAtute  Of  Medicine  Report  

Page 11: Reproducible,Research,with, Evidence5based,DataAnalysis,rpeng/talks/pengslides_transscience.pdf · Reproducible,Research,with, Evidence5based,DataAnalysis, Roger,D.,Peng,,PhD, Departmentof)Biosta/s/cs)

The  IOM  Report  

In  the  Discovery/Test  ValidaAon  stage  of  omics-­‐based  tests:  

•  Data/metadata  used  to  develop  test  should  be  made  publicly  available  

•  The  computer  code  and  fully  specified  computaAonal  procedures  used  for  development  of  the  candidate  omics-­‐based  test  should  be  made  sustainably  available  

•  “Ideally,  the  computer  code  that  is  released  will  encompass  all  of  the  steps  of  computa(onal  analysis,  including  all  data  preprocessing  steps,  that  have  been  described  in  this  chapter.  All  aspects  of  the  analysis  need  to  be  transparently  reported.”  

 

Page 12: Reproducible,Research,with, Evidence5based,DataAnalysis,rpeng/talks/pengslides_transscience.pdf · Reproducible,Research,with, Evidence5based,DataAnalysis, Roger,D.,Peng,,PhD, Departmentof)Biosta/s/cs)

What  is  Reproducible  Research?  

Published  ArAcle  

Protocol  

ScienAfic  QuesAon  

Nature  

Author  

Reader  

Express  train  to  nature  

Page 13: Reproducible,Research,with, Evidence5based,DataAnalysis,rpeng/talks/pengslides_transscience.pdf · Reproducible,Research,with, Evidence5based,DataAnalysis, Roger,D.,Peng,,PhD, Departmentof)Biosta/s/cs)

What  is  Reproducible  Research?  

Measured  Data  

AnalyAc  Data  

ComputaAonal  Results   Tables  

Figures  

Numerical  Summaries   Text  

Author  

Reader  

Processing  code   AnalyAc  code  

PresentaAon  code  

Protocol  

Nature  

Published  ArAcle  

ScienAfic  QuesAon  

Page 14: Reproducible,Research,with, Evidence5based,DataAnalysis,rpeng/talks/pengslides_transscience.pdf · Reproducible,Research,with, Evidence5based,DataAnalysis, Roger,D.,Peng,,PhD, Departmentof)Biosta/s/cs)

What’s  Next  for  Reproducibility?  

•  Reproducibility  is  criAcal  for  communica/ng  a  data  analysis  

•  One  cannot  sufficiently  describe  an  analysis  in  journal  pages  or  supplementary  materials  

•  General  consensus  about  its  importance  •  No  credible  plan  (yet)  for  how  to  implement  such  a  requirement  (hint:  this  is  what’s  next)  

Page 15: Reproducible,Research,with, Evidence5based,DataAnalysis,rpeng/talks/pengslides_transscience.pdf · Reproducible,Research,with, Evidence5based,DataAnalysis, Roger,D.,Peng,,PhD, Departmentof)Biosta/s/cs)

ReplicaAon  and  Reproducibility  

Replica(on  •  Focuses  on  the  validity  of  the  

scien/fic  claim  •  “Is  this  claim  true?”  •  The  ulAmate  standard  for  

strengthening  scienAfic  evidence  •  New  invesAgators,  data,  

analyAcal  methods,  laboratories,  instruments,  etc.  

•  ParAcularly  important  in  studies  that  can  impact  broad  policy  or  regulatory  decisions  

Reproducibility  •  Presumably  focuses  on  the  

validity  of  the  data  analysis  •  “Can  we  trust  this  analysis?”  •  Arguably  a  minimum  

standard  for  any  scienAfic  study  

•  New  invesAgators,  same  data,  same  methods  

•  Important  when  replicaAon  is  impossible  

Page 16: Reproducible,Research,with, Evidence5based,DataAnalysis,rpeng/talks/pengslides_transscience.pdf · Reproducible,Research,with, Evidence5based,DataAnalysis, Roger,D.,Peng,,PhD, Departmentof)Biosta/s/cs)

What  Problem  Does  Reproducibility  Solve?  

What  we  get  •  Transparency  •  Data  Availability  •  Sodware  /  Methods  

Availability  •  Improved  Transfer  of  

Knowledge  

What  we  do  not  get  •  Validity  /  Correctness  of  the  

analysis  

 

An  analysis  can  be  reproducible  and  s/ll  be  wrong  

Does  requiring  reproducibility  deter  bad  analysis?  

We  want  to  know  “can  we  trust  this  analysis?”  

Page 17: Reproducible,Research,with, Evidence5based,DataAnalysis,rpeng/talks/pengslides_transscience.pdf · Reproducible,Research,with, Evidence5based,DataAnalysis, Roger,D.,Peng,,PhD, Departmentof)Biosta/s/cs)

Problems  with  Reproducibility  

The  premise  of  reproducible  research  is  that  with  data/code  available,  people  can  check  each  other  and  the  whole  system  is  self-­‐correcAng  •  Addresses  the  most  “downstream”  aspect  of  the  research  process  –  post-­‐publicaAon  

•  Assumes  everyone  is  capable  of  doing  analysis  and  wants  to  achieve  the  same  goals  (i.e.  scienAfic  discovery,  search  for  truth)  

Page 18: Reproducible,Research,with, Evidence5based,DataAnalysis,rpeng/talks/pengslides_transscience.pdf · Reproducible,Research,with, Evidence5based,DataAnalysis, Roger,D.,Peng,,PhD, Departmentof)Biosta/s/cs)

An  Analogy  from  Allergic  Asthma  

Allergen  Exposure  

Specific  IgE  +  sensiAzed  mast  

cells  

Inflammatory  mediators  (histamine,  leukotrienes)  

VasodilaAon,  mucous  (symptoms),  bronchoconstricAon  

Environmental  IntervenAon   AnA-­‐IgE   AnA-­‐

inflammatory  

MedicaAon  

AnA-­‐mediator  

Page 19: Reproducible,Research,with, Evidence5based,DataAnalysis,rpeng/talks/pengslides_transscience.pdf · Reproducible,Research,with, Evidence5based,DataAnalysis, Roger,D.,Peng,,PhD, Departmentof)Biosta/s/cs)

ScienAfic  DisseminaAon  Process  

Research  conducted  

(problemaAc?)  

Paper  submiied  to  

journal  

Paper  publicaAon  

Post-­‐publicaAon  review  

Editor’s  judgment  

“MedicaAon”  

Peer  review   Reproducible  research  

Page 20: Reproducible,Research,with, Evidence5based,DataAnalysis,rpeng/talks/pengslides_transscience.pdf · Reproducible,Research,with, Evidence5based,DataAnalysis, Roger,D.,Peng,,PhD, Departmentof)Biosta/s/cs)

ScienAfic  DisseminaAon  Process  

Research  conducted  

(problemaAc?)  

Paper  submiied  to  

journal  

Paper  publicaAon  

Post-­‐publicaAon  review  

???   Editor’s  judgment  

“MedicaAon”  

Peer  review   Reproducible  research  

Page 21: Reproducible,Research,with, Evidence5based,DataAnalysis,rpeng/talks/pengslides_transscience.pdf · Reproducible,Research,with, Evidence5based,DataAnalysis, Roger,D.,Peng,,PhD, Departmentof)Biosta/s/cs)

Biostatistics (2009), 10, 3, pp. 409–423doi:10.1093/biostatistics/kxp010Advance Access publication on April 17, 2009

Air pollution and health in Scotland: a multicity study

DUNCAN LEE∗, CLAIRE FERGUSON

Department of Statistics, University of Glasgow, Glasgow, G12 8QQ [email protected]

RICHARD MITCHELL

Public Health and Health Policy, University of Glasgow, Glasgow, G12 8QQ UK

SUMMARY

This paper presents an epidemiological study investigating the effects of long-term air pollution exposureon public health in Scotland, focusing on the 4 major urban areas, Aberdeen, Dundee, Edinburgh, andGlasgow. In particular, the associations between respiratory hospital admissions in 2005 and exposure toboth PM10 and NO2 between 2002 and 2004 are estimated using a small-area ecological design. The im-plementation of such studies requires careful consideration of a number of statistical issues, including howto model spatial correlation, identifiability of the model parameters, and the possible effects of ecologicalbias. The results show that long-term exposures (over 3 years) to PM10 and NO2 are significantly associ-ated with respiratory hospital admissions in Edinburgh and Glasgow, whereas the risks for Aberdeen andDundee are generally positive but nonsignificant.

Keywords: Air pollution and health; Bayesian spatial modeling; Ecological bias.

1. INTRODUCTION

The adverse effects on health associated with air pollution exposure came to public prominence in themid 1900s as a result of high air-pollution episodes that caused large numbers of excess deaths (Ministryof Public Health, 1954; Firket, 1936). Since then, numerous studies investigating both the short- and thelong-term effects of various constituents of air pollution have been conducted, with the majority findingpositive associations between exposure and both mortality and morbidity events. Short-term studies aretypically based on an ecological (at the population level) time series design, in which counts of mortalityor morbidity events on a given day are related to pollution exposures on the preceding few days. Examplesinclude the National Morbidity, Mortality, and Air Pollution Study (Dominici and others, 2002) and AirPollution and Health: a European Approach (Katsouyanni and others, 2001), which are multicity studiesbased on America and Europe, respectively.

Long-term studies estimate the cumulative effects on health of exposure over a number of years and arebased on either individual or ecological designs. Individual-level studies relate to a large cohort of peoplewhose health status is periodically assessed and subsequently related to ambient pollution concentrations

∗To whom correspondence should be addressed.

c⃝ The Author 2009. Published by Oxford University Press. All rights reserved. For permissions, please e-mail: [email protected].

by guest on January 5, 2011biostatistics.oxfordjournals.org

Dow

nloaded from

Reproducible  Research  at  Biosta/s/cs  Biostatistics (2009), 10, 4, pp. 756–772doi:10.1093/biostatistics/kxp029Advance Access publication on July 27, 2009

Second-order estimating equations for the analysis ofclustered current status data

RICHARD J. COOK∗, DAVID TOLUSSO

Department of Statistics and Actuarial Science, University of Waterloo,Waterloo, ON, Canada N2L 3G1

[email protected]

SUMMARY

With clustered event time data, interest most often lies in marginal features such as quantiles or proba-bilities from the marginal event time distribution or covariate effects on marginal hazard functions. Cop-ula models offer a convenient framework for modeling. We present methods of estimating the baselinemarginal distributions, covariate effects, and association parameters for clustered current status data basedon second-order generalized estimating equations. We examine the efficiency gains realized from usingsecond-order estimating equations compared with first-order equations, issues of copula misspecification,and apply the methods to motivating studies including one on the incidence of joint damage in patientswith psoriatic arthritis.

Keywords: Current status data; Generalized estimating equations; Piecewise constant hazards; Relative efficiency.

1. INTRODUCTION

Current status data, also referred to as type I interval censored data, arise when the status of a subjectis only known at a single point in time, and hence, the underlying failure time is either left or rightcensored. Ayer and others (1955) show how to obtain the nonparametric maximum likelihood estimateof the distribution function with current status data using techniques from isotonic regression (Barlowand others, 1972). Methods for fitting multiplicative semiparametric models are described by Xuand others (2004) via sieve estimation, and generalized additive models are fitted using isotonic regres-sion in Shiboski (1998). Sun (2006) provides a comprehensive summary of modern statistical methods forcurrent status data.

Several authors have investigated estimation of joint models for bivariate current status data. In thespirit of Shih and Louis (1995), Wang and Ding (2000) describe a 2-stage semiparametric estimationprocedure for bivariate current status data under a Clayton copula and assess the asymptotic and empiricalproperties of the resulting estimator of Kendall’s τ. van der Laan and Jewell (2002) discuss identifiabil-ity issues for bivariate current status data and develop an expectation-maximization algorithm that canbe used to estimate associated marginal features for constructing tests of independence. Extensions ofthis work appear in Jewell and others (2005) where other topics including goodness of fit of the copula

∗To whom correspondence should be addressed.

c⃝ The Author 2009. Published by Oxford University Press. All rights reserved. For permissions, please e-mail: [email protected].

by guest on January 5, 2011biostatistics.oxfordjournals.org

Dow

nloaded from

Page 22: Reproducible,Research,with, Evidence5based,DataAnalysis,rpeng/talks/pengslides_transscience.pdf · Reproducible,Research,with, Evidence5based,DataAnalysis, Roger,D.,Peng,,PhD, Departmentof)Biosta/s/cs)

Who  Reproduces  Research?  

•  Someone  needs  to  do  something  – Re-­‐run  the  analysis;  check  results  match  – Check  the  code  for  bugs/errors  – Try  alternate  approaches;  check  sensiAvity  

•  The  need  for  someone  to  do  something  is  inherited  from  tradiAonal  noAon  of  replicaAon  

•  Who  is  “someone”  and  what  are  their  goals?  

Page 23: Reproducible,Research,with, Evidence5based,DataAnalysis,rpeng/talks/pengslides_transscience.pdf · Reproducible,Research,with, Evidence5based,DataAnalysis, Roger,D.,Peng,,PhD, Departmentof)Biosta/s/cs)

Who  Reproduces  Research?  

The  truth  is  A  

I  don’t  care  

The  truth  is  B  

The  truth  is  not  A  

Original  InvesAgator   Reproducers  

The  truth  is  A  ScienAsts  

General  Public  

???  

Page 24: Reproducible,Research,with, Evidence5based,DataAnalysis,rpeng/talks/pengslides_transscience.pdf · Reproducible,Research,with, Evidence5based,DataAnalysis, Roger,D.,Peng,,PhD, Departmentof)Biosta/s/cs)

The  Story  So  Far  

•  Reproducibility  brings  transparency,  improves  communicaAon,  and  speeds  knowledge  transfer  

•  A  lot  of  discussion  about  how  to  get  people  to  share  data  

•  Key  quesAon  of  “can  we  trust  this  analysis?”  is  not  addressed  by  reproducibility  

•  Reproducibility  addresses  potenAal  problems  long  ader  they’ve  occurred  (“downstream”)  

•  Secondary  analyses  are  inevitably  colored  by  the  interests/moAvaAons  of  others  

Page 25: Reproducible,Research,with, Evidence5based,DataAnalysis,rpeng/talks/pengslides_transscience.pdf · Reproducible,Research,with, Evidence5based,DataAnalysis, Roger,D.,Peng,,PhD, Departmentof)Biosta/s/cs)

Evidence-­‐based  Data  Analysis  

•  Most  data  analyses  involve  stringing  together  many  different  tools  and  methods  

•  Some  methods  may  be  standard  for  a  given  field,  but  others  are  oden  applied  ad  hoc  

•  We  should  apply  thoroughly  studied  (via  staAsAcal  research),  mutually  agreed  upon  methods  to  analyze  data  whenever  possible  

•  There  should  be  evidence  to  jusAfy  the  applicaAon  of  a  given  method  

Page 26: Reproducible,Research,with, Evidence5based,DataAnalysis,rpeng/talks/pengslides_transscience.pdf · Reproducible,Research,with, Evidence5based,DataAnalysis, Roger,D.,Peng,,PhD, Departmentof)Biosta/s/cs)

 IncorporaAng  StaAsAcal  ExperAse  Into  Sodware  (1991  ediAon)  

“Throughout  American  or  even  global  industry,  there  is  much  advocacy  of  staAsAcal  process  control  and  of  understanding  processes.  Sta(s(cians  have  a  process  they  espouse  but  do  not  know  anything  about.  It  is  the  process  of  pulng  together  many  Any  pieces,  the  process  called  data  analysis,  and  is  not  really  understood.”      -­‐-­‐Daryl  Pregibon,  NRC  Report  1991  

Page 27: Reproducible,Research,with, Evidence5based,DataAnalysis,rpeng/talks/pengslides_transscience.pdf · Reproducible,Research,with, Evidence5based,DataAnalysis, Roger,D.,Peng,,PhD, Departmentof)Biosta/s/cs)

Consider  the  Following  Problem  

You  have  a  sample  of  300  observaAons  on  a  conAnuous  morbidity  outcome  Y  and  a  predictor  X  that  is  conAnuous  biomarker.    Describe  a  process  to  determine  the  magnitude  of  the  linear  associa(on  between  X  and  Y  

Page 28: Reproducible,Research,with, Evidence5based,DataAnalysis,rpeng/talks/pengslides_transscience.pdf · Reproducible,Research,with, Evidence5based,DataAnalysis, Roger,D.,Peng,,PhD, Departmentof)Biosta/s/cs)

Evidence-­‐based  Histogram  Bin  Width  

Histogram of x

x

Frequency

−2 −1 0 1 2 3

05

1015

20

Sturges  HA  (1926),  JASA  Scoi  DW  (1979),  Biometrika  

Page 29: Reproducible,Research,with, Evidence5based,DataAnalysis,rpeng/talks/pengslides_transscience.pdf · Reproducible,Research,with, Evidence5based,DataAnalysis, Roger,D.,Peng,,PhD, Departmentof)Biosta/s/cs)

What  is  Data  Analysis?  

Raw  Data   Cleaning  /  ValidaAon   Pre-­‐processing   Exploratory  

data  analysis  

StaAsAcal  model  development  

SensiAvity  analysis  

Finalize  results  /  report  

StaAsAcs!  

Page 30: Reproducible,Research,with, Evidence5based,DataAnalysis,rpeng/talks/pengslides_transscience.pdf · Reproducible,Research,with, Evidence5based,DataAnalysis, Roger,D.,Peng,,PhD, Departmentof)Biosta/s/cs)

Evidence-­‐based  Data  Analysis  

•  Create  analyAc  pipelines  from  evidence-­‐based  components  –  standardize  it  

•  A  Determinis/c  Sta/s/cal  Machine  hip://goo.gl/Qvlhuv  

•  Once  an  evidence-­‐based  analyAc  pipeline  is  established,  we  shouldn’t  mess  with  it  – Analysis  with  a  “transparent  box”  

•  Reduce  the  “researcher  degrees  of  freedom”  •  Analogous  to  a  pre-­‐specified  clinical  trial  protocol  

Page 31: Reproducible,Research,with, Evidence5based,DataAnalysis,rpeng/talks/pengslides_transscience.pdf · Reproducible,Research,with, Evidence5based,DataAnalysis, Roger,D.,Peng,,PhD, Departmentof)Biosta/s/cs)

DeterminisAc  StaAsAcal  Machine  

Dataset  

Input  metadata  

Preprocessing  

Model  Filng  

SensiAvity  analysis  

Report  

Output  parameters  

Public  repository  

Benchmark  dataset  

Benchmark  dataset  

Benchmark  dataset  

Methods  secAon  

Page 32: Reproducible,Research,with, Evidence5based,DataAnalysis,rpeng/talks/pengslides_transscience.pdf · Reproducible,Research,with, Evidence5based,DataAnalysis, Roger,D.,Peng,,PhD, Departmentof)Biosta/s/cs)

Case  Study:  EsAmaAng  Acute  Effects  of  Ambient  Air  PolluAon  Exposure  

•  Acute/short-­‐term  effects  typically  esAmated  via  panel  studies  or  Ame  series  studies  

•  Work  originated  in  late  1970s  early  1980s  •  Key  quesAon:  “Are  short-­‐term  changes  in  polluAon  associated  with  short-­‐term  changes  in  a  populaAon  health  outcome?”  

•  Studies  usually  conducted  at  community  level  •  Long  history  of  staAsAcal  research  invesAgaAng  proper  methods  of  analysis  

Page 33: Reproducible,Research,with, Evidence5based,DataAnalysis,rpeng/talks/pengslides_transscience.pdf · Reproducible,Research,with, Evidence5based,DataAnalysis, Roger,D.,Peng,,PhD, Departmentof)Biosta/s/cs)

New  York  Data  

●●

●●

●●

●●

●●●

●●

●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●●●●

●●

●●●

●●

●●

●●●

●●●

●●●

●●

●●●

●●●

●●

●●

●●

●●

●●●●

●●●

●●

●●

●●

●●●●

●●

●●

●●

●●

●●

●●

●●●●

●●

●●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●●

●●

●●●●

●●

●●●

●●●

●●●●

●●●

●●●

●●

●●

●●●

●●●●

●●

●●

●●

●●

●●

●●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●●●

●●

●●

●●●●

●●

●●

●●

●●

●●●

●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●●

●●

●●

●●

●●

●●●

●●

●●●

●●●

●●●●

●●

●●●●

●●

●●

●●●●

●●●

●●

●●

●●

●●●

●●

●●●

●●

●●

●●

●●

●●●●

●●●

●●●●

●●

●●

●●

●●●

●●

●●

●●

●●

●●●●

●●

●●●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●

●●●

●●

●●

●●

●●

●●

●●●

●●

●●

●●●

●●

●●

●●

●●●

●●●●

●●

●●

●●

●●

●●

●●

●●●●

●●

●●

●●

●●●●

●●●

●●

●●

●●

●●

●●

●●●●●

●●

●●

●●

●●●

●●

●●●

●●

●●

●●

●●

●●●●●

●●

●●

●●

2002 2003 2004 2005

120

160

200

Mor

talit

y co

unt

●●●●●●

●●●●

●●

●●●

●●

●●

●●

●●

●●●

●●●●

●●

●●

●●

●●

●●●●●●●●

●●●●●●

●●●

●●

●●

●●

●●

●●

●●

●●●●●●●●●

●●

●●

●●

●●●

●●●●

●●

●●●

●●

●●

●●

●●●●

●●●●

●●●●●●

●●●●●

2002 2003 2004 2005

020

4060

PM10

Page 34: Reproducible,Research,with, Evidence5based,DataAnalysis,rpeng/talks/pengslides_transscience.pdf · Reproducible,Research,with, Evidence5based,DataAnalysis, Roger,D.,Peng,,PhD, Departmentof)Biosta/s/cs)

Case  Study:  EsAmaAng  Acute  Effects  of  Ambient  Air  PolluAon  Exposure  

•  Can  we  encode  everything  that  we  have  found  in  staAsAcal/epidemiological  research  into  a  single  package?  

•  Time  series  studies  do  not  have  a  huge  range  of  variaAon;  typically  involves  similar  types  of  data  and  similar  quesAons  

•  We  can  create  a  determinisAc  staAsAcal  machine  for  this  area?      

Page 35: Reproducible,Research,with, Evidence5based,DataAnalysis,rpeng/talks/pengslides_transscience.pdf · Reproducible,Research,with, Evidence5based,DataAnalysis, Roger,D.,Peng,,PhD, Departmentof)Biosta/s/cs)

DSM  Modules  for  Time  Series  Studies  of  Air  PolluAon  and  Health  

1.  Check  for  outliers,  high  leverage,  overdispersion  2.  Fill  in  missing  data?  NO!  3.  Model  selecAon:  EsAmate  degrees  of  freedom  

to  adjust  for  unmeasured  confounders  –  Other  aspects  of  model  not  as  criAcal  

4.  MulAple  lag  analysis  5.  SensiAvity  analysis  wrt  –  Unmeasured  confounder  adjustment  –  InfluenAal  points  

Dominici,  McDermoi,  HasAe  (2004)  JASA;  Peng,  Dominici,  Louis  (2006)  JRSS-­‐A    

Page 36: Reproducible,Research,with, Evidence5based,DataAnalysis,rpeng/talks/pengslides_transscience.pdf · Reproducible,Research,with, Evidence5based,DataAnalysis, Roger,D.,Peng,,PhD, Departmentof)Biosta/s/cs)

DSM  Output  

Page 37: Reproducible,Research,with, Evidence5based,DataAnalysis,rpeng/talks/pengslides_transscience.pdf · Reproducible,Research,with, Evidence5based,DataAnalysis, Roger,D.,Peng,,PhD, Departmentof)Biosta/s/cs)

Where  to  Go  From  Here?  

•  One  DSM  is  not  enough,  we  need  many!  •  Different  problems  warrant  different  approaches  and  experAse  

•  A  curated  library  of  machines  providing  state-­‐of-­‐the  art  analysis  pipelines  

•  A  CRAN/CPAN/CTAN/…  for  data  analysis  •  Or  a  “Cochrane  CollaboraAon”  for  data  analysis  

Page 38: Reproducible,Research,with, Evidence5based,DataAnalysis,rpeng/talks/pengslides_transscience.pdf · Reproducible,Research,with, Evidence5based,DataAnalysis, Roger,D.,Peng,,PhD, Departmentof)Biosta/s/cs)

Cochrane  CollaboraAon  

Page 39: Reproducible,Research,with, Evidence5based,DataAnalysis,rpeng/talks/pengslides_transscience.pdf · Reproducible,Research,with, Evidence5based,DataAnalysis, Roger,D.,Peng,,PhD, Departmentof)Biosta/s/cs)

Cochrane  CollaboraAon  

Page 40: Reproducible,Research,with, Evidence5based,DataAnalysis,rpeng/talks/pengslides_transscience.pdf · Reproducible,Research,with, Evidence5based,DataAnalysis, Roger,D.,Peng,,PhD, Departmentof)Biosta/s/cs)

A  Curated  Library  of  Data  Analysis  

•  Provide  packages  that  encode  data  analysis  pipelines  for  given  problems,  technologies,  quesAons  

•  Curated  by  experts  knowledgeable  in  the  field  •  DocumentaAon/references  given  supporAng  each  module  in  the  pipeline  

•  Changes  introduced  ader  passing  relevant  benchmarks/unit  tests  

Page 41: Reproducible,Research,with, Evidence5based,DataAnalysis,rpeng/talks/pengslides_transscience.pdf · Reproducible,Research,with, Evidence5based,DataAnalysis, Roger,D.,Peng,,PhD, Departmentof)Biosta/s/cs)

Summary  

•  Reproducible  research  is  important,  but  does  not  necessarily  solve  the  criAcal  quesAon  of  whether  a  data  analysis  is  trustworthy  

•  Reproducible  research  focuses  on  the  most  “downstream”  aspect  of  research  disseminaAon  

•  Evidence-­‐based  data  analysis  would  provide  standardized,  best  pracAces  for  given  scienAfic  areas  and  quesAons  

•  Gives  reviewers  an  important  tool  without  dramaAcally  increasing  the  burden  on  them  

•  More  effort  should  be  put  into  improving  the  quality  of  “upstream”  aspects  of  scienAfic  research  

Page 42: Reproducible,Research,with, Evidence5based,DataAnalysis,rpeng/talks/pengslides_transscience.pdf · Reproducible,Research,with, Evidence5based,DataAnalysis, Roger,D.,Peng,,PhD, Departmentof)Biosta/s/cs)

Coming  This  Fall!  G U E S T E D I T O R ’ S I N T R O D U C T I O N

COMPUTING IN SCIENCE & ENGINEERING 11

Reproducible Research: Tools and Strategies for Scientific Computing

T his special issue is comprised of ar-ticles contributed by participants in a workshop held in Vancouver in July 2011 called “Reproducible Research:

Tools and Strategies for Scientific Computing.” The workshop was co-organized by Randall J. LeVeque (the Founders’ Term Professor of Ap-plied Mathematics at the University of Wash-ington), Ian M. Mitchell (an associate professor of computer science at the University of British Columbia), and myself. One of our aims was to improve the visibility of the nascent group of tool

builders working to facilitate really reproducible research in computational science.

This special issue focuses on tools and strate-gies for reproducible computational science and includes seven articles, each describing software development efforts. It begins with an introduc-tory article by the three workshop co-organizers

1521-9615/12/$31.00 © 2012 IEEECOPUBLISHED BY THE IEEE CS AND THE AIP

Victoria StoddenColumbia University

CISE-14-4-Gei.indd 11 6/13/12 2:39 PM

Page 43: Reproducible,Research,with, Evidence5based,DataAnalysis,rpeng/talks/pengslides_transscience.pdf · Reproducible,Research,with, Evidence5based,DataAnalysis, Roger,D.,Peng,,PhD, Departmentof)Biosta/s/cs)

Johns  Hopkins  Data  Science  

Page 44: Reproducible,Research,with, Evidence5based,DataAnalysis,rpeng/talks/pengslides_transscience.pdf · Reproducible,Research,with, Evidence5based,DataAnalysis, Roger,D.,Peng,,PhD, Departmentof)Biosta/s/cs)

Johns  Hopkins  Data  Science  

•  All  online,  all  the  Ame  •  Probability/math  stat,  staAsAcal  inference  •  Gelng  +  cleaning  data  •  R  programming  •  Regression  modeling  •  PredicAon  +  machine  learning  •  Exploratory  data  analysis  •  Reproducible  research  •  Data  products  •  Capstone  project  with  industry  partners