Performance: Test cases - United Nations Economic Commission for

Download Report

Transcript Performance: Test cases - United Nations Economic Commission for

R in the Statistical Office
R in the Statistical Office:
The UNIDO Experience
Valentin Todorov
UNIDO
[email protected]
MSIS 2010 (Daejeon, 26-29 April 2010)
26.4.2010
MSIS 2010, Daejeon: Valentin Todorov
1
R in the Statistical Office
Outline
• Introduction: the R Platform and Availability
• R Extensibility (R Packages)
• R in UNIDO statistical process: three examples
– R as a Mediator (R Interfaces)
– R as a Graphics Engine (R, LaTeX and Sweave)
– Nowcasting tool for the Manufacturing Value Added (MVA)
• Summary and Conclusions
26.4.2010
MSIS 2010, Daejeon: Valentin Todorov
2
R in the Statistical Office
What is R: Platform
• R is “a system for statistical computation and graphics. It
provides, among other things, a programming language,
high-level graphics, interfaces to other languages and
debugging facilities”
• Developed after the S language and environment
– S was developed at Bell Labs (John Chambers et al.)
– S-Plus: a value added implementation of the S language- Insightful
Corporation
– much code written for S runs unaltered under R
• Significantly influenced by Scheme, a Lisp dialect
26.4.2010
MSIS 2010, Daejeon: Valentin Todorov
3
R in the Statistical Office
What is R: History
• Ihaka and Gentleman, University of Auckland (New
Zealand)
– 1993 a preliminary version of R
– 1995 released under the GNU Public License
– Now: R-core team consisting of 17 members including John
Chambers
• R provides a wide variety of statistical (linear and nonlinear modelling, classical statistical tests, time-series
analysis, classification, clustering, robust methods and
many more) and graphical techniques
26.4.2010
MSIS 2010, Daejeon: Valentin Todorov
4
R in the Statistical Office
What is R: Availability
• R is available as Free Software under the terms of the
GNU General Public License (GPL)
• R is available for:
– wide variety of UNIX platforms (including FreeBSD and Linux)
– Windows
– MacOS
• Add-on functionality is available in the form of packages
from CRAN: http://cran.r-project.org/
26.4.2010
MSIS 2010, Daejeon: Valentin Todorov
5
R in the Statistical Office
R Extensibility (R Packages)
• One of the most important features of R is its extensibility
by creating packages of functions and data.
• The R package system provides a framework for
developing, documenting, and testing extension code.
• Packages can include R code, documentation, data and
foreign code written in C or Fortran.
• Packages are distributed through the CRAN repository –
http://cran.r-project.org - currently more than 1300
packages covering a wide variety of statistical methods
and algorithms. ‘base’ and ‘recommended’ packages are
included in all binary distributions.
26.4.2010
MSIS 2010, Daejeon: Valentin Todorov
6
R in the Statistical Office
26.4.2010
MSIS 2010, Daejeon: Valentin Todorov
7
R in the Statistical Office
I. R as a mediator (R Interfaces)
• Using a statistical system is not done in isolation
• Import data for analysis
• Export data for further processing
 use the right tool for the right work
• Export results for report writing
• Even in a small research department (UNIDO):
SAS, Stata, Eviews, Octave, SPSS and R user
26.4.2010
MSIS 2010, Daejeon: Valentin Todorov
8
R in the Statistical Office
R as a mediator (R Interfaces)
• Reading and writing data
(text files, XML, spreadsheet
like data, e.g. Excel
• Read and write data formats
of SAS, S-Plus, SPSS,
EpiInfo, STATA, SYSTAT,
Octave – package foreign
• Emulation of Matlab –
package matlab
26.4.2010
MSIS 2010, Daejeon: Valentin Todorov
9
R in the Statistical Office
R as a mediator: the foreign package
library(foreign)
df <- read.dbf(“myfile.dbf")
# DBase
df <- read.epiinfo("myfile.epiinfo")
# Epi Info
df <- read.mtp("myfile.mtp") # Minitab portable worksheet
df <- read.octave("myfile.octave")
# Octave
df <- read.ssd("myfile.ssd")
# SAS version 6
df <- read.xport("myfile.xport")
# SAS XPORT file
df <- read.spss("myfile.sav")
# SPSS
df <- read.dta("myfile.dta")
# Stata
df <- read.systat("myfile.sys")
# Systat
26.4.2010
MSIS 2010, Daejeon: Valentin Todorov
10
R in the Statistical Office
R as a mediator (Accessing data on the Internet)
• Reading data from an URL:
– readLines() to read arbitrary text
– read.table() to read a file with observations and variables (first line
can be used for variable names)
– read.csv() to read comma separated values.
• Example (from Kleinman and Horton, 2009)
ch <- url("http://www.math.smith.edu/sasr/testdata")
df <- readLines(ch)
##
df <- read.table("http://www.math.smith.edu/sasr/testdata")
##
df <- read.csv("http://www.math.smith.edu/sasr/file.csv")
26.4.2010
MSIS 2010, Daejeon: Valentin Todorov
11
R in the Statistical Office
R as a mediator (XML processing)
• Use package XML
– xmlRoot() to open the connection to the file
– xmlSApply() and xmlValue() are called recursively to process the
file.
– A character matrix is returned: columns correspond to observations
and rows correspond to variables.
• Example (from Kleinman and Horton, 2009)
library(XML)
surl <- http://www.math.smith.edu/sasr/datasets/help.xml
doc <- xmlRoot(xmlTreeParse(surl ))
tmp <- xmlSApply(doc, function(x) xmlSApply(x, xmlValue))
df <- t(tmp)[,-1]
26.4.2010
MSIS 2010, Daejeon: Valentin Todorov
12
R in the Statistical Office
SDMX example: Retrieve IMF/IFS data
library(XML)
surl <- "c:/download/Exrate4Unido.xml"
doc <- as.list(xmlRoot(xmlTreeParse(surl)))
## Get the data for Korea
kr <- doc[[which(xmlSApply(doc, function(x) xmlAttrs(x)[3]) == "Korea")]]
xmlAttrs(kr)
Frequency
"A"
CountryName
"Korea"
TS_Key
"542..RF.ZF..."
Units
Database
"IFS"
Country
"542"
Descriptor
"MARKET RATE, PERIOD AVERAGE"
Scale
"National Currency per US Dollar“
26.4.2010
MSIS 2010, Daejeon: Valentin Todorov
"None"
13
R in the Statistical Office
SDMX example: Retrieve IMF/IFS data (2)
getExdata <- function(x) { out <<- rbind(out,as.numeric(xmlAttrs(x))) }
out <- data.frame()
xmlSApply(kr, getExdata)
out
…
27 1974 404.4725
28 1975 484.0000
29 1976 484.0000
30 1977 484.0000
31 1978 484.0000
32 1979 484.0000
33 1980 607.4325
34 1981 681.0283
…
26.4.2010
MSIS 2010, Daejeon: Valentin Todorov
14
R in the Statistical Office
R as a mediator (Databases)
• Communication with RDBMS – ROracle, RMySql,
RSQLite, RmSQL, RPgSQL, RODBC – large data sets,
concurrency
• Package filehash – a simple key-value style database, the
data are stored on disk but are handled like data sets
• Can use compiled native code in C, C++, Fortran, Java
26.4.2010
MSIS 2010, Daejeon: Valentin Todorov
15
R in the Statistical Office
R as a mediator: IDSB Example
• Industrial Demand and Supply Balance (IDSB)
Database - data sets based on ISIC Rev.3 at 4-digit level
• Contains annual time series data (in current US dollars) for
eight interrelated items
• Data are derived from:
– INDSTAT: Output data reported by National Statistical Offices
– COMTRADE: UNIDO estimates for ISIC-based international trade
data
• A new, related to IDSB product will contain also Index of
Industrial Production data (UNSD)
26.4.2010
MSIS 2010, Daejeon: Valentin Todorov
16
R in the Statistical Office
R as a mediator: IDSB Example (2)
• The generation of the final data set involves
– combination of two independent data sets (INDSTAT and
COMTRADE),
– conversion from one classification (SITC) to another (ISIC),
– conversion of the monetary values from current national currency
to current USD and other minor adjustments of the data.
• Each single data set is verified thoroughly and its quality is
guaranteed
• But the verification of the synthesized data set is a serious
challenge for the statistical staff of the Unit:
A comprehensive screening data set is created 
26.4.2010
MSIS 2010, Daejeon: Valentin Todorov
17
R in the Statistical Office
R as a mediator: IDSB Example (3)
26.4.2010
MSIS 2010, Daejeon: Valentin Todorov
18
R in the Statistical Office
R as a mediator: IDSB Example: R code
## First load the RODBC library. If not yet installed, install it using
##
install.packages("RODBC")
##
library(RODBC)
## Open the ODBC connection to the MDB file ’fname’
##
ch <- odbcConnectAccess("C:/work/idsb34screen.mdb")
## Create an SQL query of the type:
##
"SELECT * FROM table_name WHERE where_condition"
## Execute the query and obtain the selected data in a dataframe
##
sql <- "Select * from idsb34 where MXMARK <>’’"
xdata <- sqlQuery(ch, sql)
26.4.2010
MSIS 2010, Daejeon: Valentin Todorov
19
R in the Statistical Office
II. R as a Graphics Engine
• A natural way to visualize data are graphs and plots
• Publication quality displays should be both informative and
aesthetically pleasing (Tufte, 2001):
– present many numbers in a small space;
– encourage the eye to compare different pieces of data
• The graphics have to be mingled with text explaining and
commenting them
• The standard approach:
POINT & CLICK – WYSIWYG – COPY & PASTE
• The proposed solution:
R + LaTeX + BibTeX => Sweave  PDF
26.4.2010
MSIS 2010, Daejeon: Valentin Todorov
20
R in the Statistical Office
The Example: International Yearbook of
Industrial Statistics
• A unique and comprehensive source of information, the
only international publication providing worldwide statistics
on performance and trends in the manufacturing sector.
• Designed to facilitate international comparisons relating to
manufacturing activity, industrial development and
performance.
• Data which can be used to analyze patterns of growth and
related long term trends, structural change and industrial
performance in individual industries.
• A new graphical section presenting the major trends of
growth and distribution of manufacturing in the world.
26.4.2010
MSIS 2010, Daejeon: Valentin Todorov
21
R in the Statistical Office
26.4.2010
MSIS 2010, Daejeon: Valentin Todorov
22
R in the Statistical Office
26.4.2010
MSIS 2010, Daejeon: Valentin Todorov
23
R in the Statistical Office
Yearbook Graphics: Requirements
• The software tool we are looking for should fulfil as a
minimum the following requirements
– Create publication quality graphics
– Interface easily with the other components of the production line
(SAS , Sybase, .Net)
– Comply with the submission guidelines of the publisher – e.g. the
final document must contain only embedded fonts.
– Provide means for easy text and image placement. Whenever the
data are changed the document should be (preferably
automatically) regenerated.
– Use the same fonts in figure labels as in the main document
– Easy to maintain and extend
26.4.2010
MSIS 2010, Daejeon: Valentin Todorov
24
R in the Statistical Office
The Components: R Graphics
• One of the most important strengths of R – simple
exploratory graphics as well as well-designed publication
quality plots
• The graphics can include mathematical symbols and
formulae where needed
• Can produce graphics in many formats:
–
–
–
–
26.4.2010
On screen
PS and PDF for including in LaTex and pdfLaTeX or for distribution
PNG or JPEG for the Web
On Windows, metafiles for Word, PowerPoint, etc
MSIS 2010, Daejeon: Valentin Todorov
25
R in the Statistical Office
R Graphics: basic and multipanel plots (trellis)
Boxplot
virginica
7.5
Petal
Length
Three
6.5
Varieties
Sepal
Width
of
4.5
5.5
Sepal.Width
0.8
0.4
0.0
Density
1.2
Histogram
2.0
2.5
3.0
3.5
4.0
Sepal
Length
setosa versicolor
Iris
Sepal.Width
setosa
Normal Q-Q Plot
Petal
Length
4.0
Petal
Length
Sepal
Width
3.0
Sepal.Width
Sepal
Length
2.0
3.0
2.0
Sepal.Width
4.0
Bagplot
versicolor
4.5
5.5
6.5
Sepal.Length
26.4.2010
7.5
-2
-1
0
1
Sepal
Width
Sepal
Length
2
norm quantiles
MSIS 2010, Daejeon: Valentin Todorov
Scatter Plot Matrix
26
R in the Statistical Office
R Graphics: parallel plot and coplot
Given : depth
Three
virginica
Petal
Length
100
200
300
400
500
600
Varieties
of
Sepal
Width
165
170
175
180
185
165
setosa
180
185
-25
Sepal
Length
versicolor
-15
lat
-35
Petal
Length
175
-15
Iris
170
-35
-25
Sepal
Width
Sepal
Length
Min
165
Max
170
175
180
185
long
26.4.2010
MSIS 2010, Daejeon: Valentin Todorov
27
R in the Statistical Office
The Components: TeX and LaTeX
• TeX: a typesetting system (computer program) for
producing nicely printed, publication quality output, freely
available: Donald E. Knuth, 1974
• LaTeX: a component designed to shield the author from
the details of TeX; Lamport (1994)
– Available for free from http://www.latex-project.org/ftp.html for
Linux, MacOs and Windows.
• BibTeX: A simple tool to create a bibliography in a LaTeX
document
– a uniform style is achieved, which easily can be replaced by
another
– a unified library of references shared among publications and
authors
26.4.2010
MSIS 2010, Daejeon: Valentin Todorov
28
R in the Statistical Office
The Components: Sweave
• A suitable tool that allows to embed the code for complete
data analysis in documents (see Leisch, 2002)
• Create dynamic reports, which can be updated
automatically if data or analysis change
• The master document (.Rnw) contains:
– the necessary programming code for obtaining of the graphs,
tables, etc. written in R
– the text written in LaTeX
• The document is run through R
– all the data analysis is performed on the fly
– the generated output - tables, graphs, etc. is inserted into the final
LaTeX document.
26.4.2010
MSIS 2010, Daejeon: Valentin Todorov
29
R in the Statistical Office
III. Nowcasting MVA for Cross-country
Comparison
• UNIDO maintains a unique industrial statistics database
INDSTAT) – updated regularly with data collected from
NSOs
• A separate database – compilation of statistics related to
MVA – growth rate and share in GDP
• Published in the International Yearbook of Industrial
Statistics and on the statistical pages of the UNIDO web
site
• For current economic analysis it is crucial that the
Yearbook presents data for the most recent years
26.4.2010
MSIS 2010, Daejeon: Valentin Todorov
30
R in the Statistical Office
Nowcasting MVA: The Model
• The database consists of annual values of MVA and GDP
at constant 2000 prices for around 200 countries
• GDP data are available up to the current year:
– For earlier years the actual GDP values are used
– For the most recent one or two years the GDP values are derived
from the nowcasts of GDP growth rates reported in the World
Economic Outlook of IMF (see Artis, 1996)
• MVA – a time-gap of at least one year: nowcasting
• MVA is strongly connected to the GDP
•  this suggests to nowcast MVA on the basis of the
estimated relationship between contemporaneous values
of MVA and GDP
26.4.2010
MSIS 2010, Daejeon: Valentin Todorov
31
R in the Statistical Office
Nowcasting MVA: The Model (2)
• We consider models based on the following general
representation of MVA
MVAi ,t  MVAi ,t 1 (1  gMVAi ,t ) )
where the MVA growth rate is modelled as
gMVAi ,t  ai  bi gGDPi ,t  ci gMVAi ,t 1  ei ,t
and ei,t is white noise.
• This general model can be specialized down to four different
models (see Boudt, Todorov and Upadhyaya, 2009)
26.4.2010
MSIS 2010, Daejeon: Valentin Todorov
32
R in the Statistical Office
Nowcasting MVA: Estimation
26.4.2010
MSIS 2010, Daejeon: Valentin Todorov
33
R in the Statistical Office
Nowcasting MVA: Estimation
• The standard OLS estimator may be biased because of
– violation of the assumption of exogeneity of the regressors with
respect to the error term
– presence of outliers in the data
• What are outliers:
– atypical observations which are inconsistent with the rest of the data
or deviate from the postulated model
– may arise through contamination, errors in data gathering, or
misspecification of the model
– classical statistical methods are very sensitive to such data
• For this reason we also consider a robust alternative to the
OLS estimator, namely the MM estimator
26.4.2010
MSIS 2010, Daejeon: Valentin Todorov
34
R in the Statistical Office
26.4.2010
MSIS 2010, Daejeon: Valentin Todorov
35
R in the Statistical Office
Nowcasting MVA: MM-Estimator
• Robust methods: produce reasonable results even when
one or more outliers may appear in the data
• The MM regression estimator is a two step estimator:
– First step – LTS (Least Trimmed Squares) – estimates the parameter
vector that minimizes the sum of the 50% smallest square residuals
– This estimate is used as a starting value for M-estimation where a
loss function is minimized that downweights outliers
• Has a high efficiency under the linear regression model with
normally distributed errors
• Because of the LTS initialization it is highly robust
• For details see Maronna et al (2006)
26.4.2010
MSIS 2010, Daejeon: Valentin Todorov
36
R in the Statistical Office
Nowcasting MVA: MM-Estimator in R
• Package robustbase:
• Provides “essential robust statistics” within R available in a
single package
• Provides tools that allow analyzing data with robust
methods:
– Regression including model selection
– Multivariate statistics
• Aims to cover the book of Maronna et al (2006)
26.4.2010
MSIS 2010, Daejeon: Valentin Todorov
37
R in the Statistical Office
Summary and Outlook
• An increasing demand for statistical tools which combine
ease of use and availability of newest analytical methods.
• Provided by the flexibility of the statistical programming
language and environment R
• Illustrated by examples from the statistical production
process of UNIDO
• Future development:
– R for survey data analysis
– Detection of outliers in survey data with R
– Imputation of missing values in multivariate data with R
26.4.2010
MSIS 2010, Daejeon: Valentin Todorov
38