Basic Civil Engineering first year Notes- Chapter 4 Building.pptx
An Empirical Analysis of Software Productivity Over Time
1. Presented at Metrics 2005 - Como, Italy
An Empirical Analysis of
Software Productivity Over Time
Rahul Premraj1 Martin Shepperd2 Barbara Kitchenham3,4
Pekka Forselius5
1 Bournemouth University, UK
2 Brunel University, UK
3 National ICT, Australia
4 Keele University, UK
5 Software Technology Transfer Finland Oy, Finland
11th IEEE Symposium on Software Metrics, 2005
Como, Italy
Premraj, Shepperd, Kitchenham, Forselius Empirical Analysis of Software Productivity
2. Presented at Metrics 2005 - Como, Italy
Contents
Contents
1 Background to the Data Set
2 Results
1 Scale Economies
2 Productivity Trends
3 Sources of Variance
3 Conclusions
Premraj, Shepperd, Kitchenham, Forselius Empirical Analysis of Software Productivity
3. Presented at Metrics 2005 - Como, Italy
Contents
Contents
1 Background to the Data Set
2 Results
1 Scale Economies
2 Productivity Trends
3 Sources of Variance
3 Conclusions
Premraj, Shepperd, Kitchenham, Forselius Empirical Analysis of Software Productivity
4. Presented at Metrics 2005 - Como, Italy
Contents
Contents
1 Background to the Data Set
2 Results
1 Scale Economies
2 Productivity Trends
3 Sources of Variance
3 Conclusions
Premraj, Shepperd, Kitchenham, Forselius Empirical Analysis of Software Productivity
5. Presented at Metrics 2005 - Como, Italy
Contents
Contents
1 Background to the Data Set
2 Results
1 Scale Economies
2 Productivity Trends
3 Sources of Variance
3 Conclusions
Premraj, Shepperd, Kitchenham, Forselius Empirical Analysis of Software Productivity
6. Presented at Metrics 2005 - Como, Italy
Contents
Contents
1 Background to the Data Set
2 Results
1 Scale Economies
2 Productivity Trends
3 Sources of Variance
3 Conclusions
Premraj, Shepperd, Kitchenham, Forselius Empirical Analysis of Software Productivity
7. Presented at Metrics 2005 - Como, Italy
Contents
Contents
1 Background to the Data Set
2 Results
1 Scale Economies
2 Productivity Trends
3 Sources of Variance
3 Conclusions
Premraj, Shepperd, Kitchenham, Forselius Empirical Analysis of Software Productivity
8. Presented at Metrics 2005 - Como, Italy
Contents
Contents
1 Background to the Data Set
2 Results
1 Scale Economies
2 Productivity Trends
3 Sources of Variance
3 Conclusions
Premraj, Shepperd, Kitchenham, Forselius Empirical Analysis of Software Productivity
9. Presented at Metrics 2005 - Como, Italy
The “Finnish Data Set”
Also known as the Experience Pro data set.
Result of commercial initiatives by Software Technology
Transfer Finland (STTF).
In total there are 622 projects and 102 features collected
including size, effort, factors characterising development
environment, target technology, etc.
Includes software projects completed in Finland between 1978
and 2003.
93% of the projects are new development projects and the
remainder are maintenance projects.
Only completed projects submitted.
Premraj, Shepperd, Kitchenham, Forselius Empirical Analysis of Software Productivity
10. Presented at Metrics 2005 - Como, Italy
The “Finnish Data Set”
Also known as the Experience Pro data set.
Result of commercial initiatives by Software Technology
Transfer Finland (STTF).
In total there are 622 projects and 102 features collected
including size, effort, factors characterising development
environment, target technology, etc.
Includes software projects completed in Finland between 1978
and 2003.
93% of the projects are new development projects and the
remainder are maintenance projects.
Only completed projects submitted.
Premraj, Shepperd, Kitchenham, Forselius Empirical Analysis of Software Productivity
11. Presented at Metrics 2005 - Como, Italy
The “Finnish Data Set”
Also known as the Experience Pro data set.
Result of commercial initiatives by Software Technology
Transfer Finland (STTF).
In total there are 622 projects and 102 features collected
including size, effort, factors characterising development
environment, target technology, etc.
Includes software projects completed in Finland between 1978
and 2003.
93% of the projects are new development projects and the
remainder are maintenance projects.
Only completed projects submitted.
Premraj, Shepperd, Kitchenham, Forselius Empirical Analysis of Software Productivity
12. Presented at Metrics 2005 - Como, Italy
The “Finnish Data Set”
Also known as the Experience Pro data set.
Result of commercial initiatives by Software Technology
Transfer Finland (STTF).
In total there are 622 projects and 102 features collected
including size, effort, factors characterising development
environment, target technology, etc.
Includes software projects completed in Finland between 1978
and 2003.
93% of the projects are new development projects and the
remainder are maintenance projects.
Only completed projects submitted.
Premraj, Shepperd, Kitchenham, Forselius Empirical Analysis of Software Productivity
13. Presented at Metrics 2005 - Como, Italy
The “Finnish Data Set”
Also known as the Experience Pro data set.
Result of commercial initiatives by Software Technology
Transfer Finland (STTF).
In total there are 622 projects and 102 features collected
including size, effort, factors characterising development
environment, target technology, etc.
Includes software projects completed in Finland between 1978
and 2003.
93% of the projects are new development projects and the
remainder are maintenance projects.
Only completed projects submitted.
Premraj, Shepperd, Kitchenham, Forselius Empirical Analysis of Software Productivity
14. Presented at Metrics 2005 - Como, Italy
The “Finnish Data Set”
Also known as the Experience Pro data set.
Result of commercial initiatives by Software Technology
Transfer Finland (STTF).
In total there are 622 projects and 102 features collected
including size, effort, factors characterising development
environment, target technology, etc.
Includes software projects completed in Finland between 1978
and 2003.
93% of the projects are new development projects and the
remainder are maintenance projects.
Only completed projects submitted.
Premraj, Shepperd, Kitchenham, Forselius Empirical Analysis of Software Productivity
15. Presented at Metrics 2005 - Como, Italy
The “Finnish Data Set”
Also known as the Experience Pro data set.
Result of commercial initiatives by Software Technology
Transfer Finland (STTF).
In total there are 622 projects and 102 features collected
including size, effort, factors characterising development
environment, target technology, etc.
Includes software projects completed in Finland between 1978
and 2003.
93% of the projects are new development projects and the
remainder are maintenance projects.
Only completed projects submitted.
Premraj, Shepperd, Kitchenham, Forselius Empirical Analysis of Software Productivity
16. Project Distribution by Business Sector
70
6%
Insurance
12%
Banking
60 Public Admin. 37% 8%
Other
Manufacturing
Retail
50
15%
Project Count
40 22%
30
20
10
0
0 ’78 ’82 ’83 ’85 ’86 ’87 ’88 ’89 ’90 ’91 ’92 ’93 ’94 ’95 ’96 ’97 ’98 ’99 ’00 ’01 ’02 ’03
Years
17. Presented at Metrics 2005 - Como, Italy
Data Editing
Of 622 projects, the following were removed:
3 projects that were not completed.
5 projects with non-standard size measurement.
Projects with implausible delivery rates (i.e. < 1FP hr −1 (6
projects) and > 30FP hr −1 (6 projects))
Thus, in total 20 projects were removed i.e. 3.2% of the data set.
Premraj, Shepperd, Kitchenham, Forselius Empirical Analysis of Software Productivity
18. Presented at Metrics 2005 - Como, Italy
Productivity and Economies of Scale
(1) Raw Data
(2) Natural Log-Scale Data
(5) Remove Projects with
(3) Build Log-Linear Model Cook’s Distance > 4/n
ln( Effort ) = a + b ln( Size)
(6) Build Log-Linear Model
ln( Effort ) = a + b ln( Size)
(4) Re-transform Data
into Original Scale
(7) Re-transform Data
Effort = a ( Size)b into Original Scale
Effort = a ( Size)b
Premraj, Shepperd, Kitchenham, Forselius Empirical Analysis of Software Productivity
19. Presented at Metrics 2005 - Como, Italy
All Projects Production Function
7
Effort vs. Size
- All Projects -
6
Power Model
Power Model - Outliers
Effort = 7.345 (Size)0.961
Outliers 0.909 < b < 1.014 and
5
Effort (Hours - ∗104)
R 2 = 0.683.
4
3
2
- Without 31 Outliers -
1
Effort = 6.13 (Size)0.993
0.94 < b < 1.047
0
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 5500
Project Size (EP20 Function Points)
Premraj, Shepperd, Kitchenham, Forselius Empirical Analysis of Software Productivity
20. Presented at Metrics 2005 - Como, Italy
All Projects Production Function
7
Effort vs. Size
- All Projects -
6
Power Model
Power Model - Outliers
Effort = 7.345 (Size)0.961
Outliers 0.909 < b < 1.014 and
5
Effort (Hours - ∗104)
R 2 = 0.683.
4
3
2
- Without 31 Outliers -
1
Effort = 6.13 (Size)0.993
0.94 < b < 1.047
0
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 5500
Project Size (EP20 Function Points)
Premraj, Shepperd, Kitchenham, Forselius Empirical Analysis of Software Productivity
21. Presented at Metrics 2005 - Como, Italy
All Projects Production Function
7
Effort vs. Size
- All Projects -
6
Power Model
Power Model - Outliers
Effort = 7.345 (Size)0.961
Outliers 0.909 < b < 1.014 and
5
Effort (Hours - ∗104)
R 2 = 0.683.
4
3
2
- Without 31 Outliers -
1
Effort = 6.13 (Size)0.993
0.94 < b < 1.047
0
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 5500
Project Size (EP20 Function Points)
Premraj, Shepperd, Kitchenham, Forselius Empirical Analysis of Software Productivity
22. Presented at Metrics 2005 - Como, Italy
Past and Present Comparison
FinnishMF Finnish602 - MF
Start dates 1978-94 1997-2003
No. of companies 26 17
No. of projects 206 401
Project sizes (FPs) 33−3375 27−5060
Productivity (FPhr−1 ) 0.177 0.233
Table: Na¨ Productivity Comparison of 1978-94 and 1997-2003
ıve
Why Na¨
ıve?
Many differences between both samples of data.
Non-constant distribution of projects across business sectors.
Maintenance projects were added only 1997 onwards.
Projects exhibited a tendency to decrease in size with time.
Premraj, Shepperd, Kitchenham, Forselius Empirical Analysis of Software Productivity
23. Presented at Metrics 2005 - Como, Italy
Past and Present Comparison
FinnishMF Finnish602 - MF
Start dates 1978-94 1997-2003
No. of companies 26 17
No. of projects 206 401
Project sizes (FPs) 33−3375 27−5060
Productivity (FPhr−1 ) 0.177 0.233
Table: Na¨ Productivity Comparison of 1978-94 and 1997-2003
ıve
Why Na¨
ıve?
Many differences between both samples of data.
Non-constant distribution of projects across business sectors.
Maintenance projects were added only 1997 onwards.
Projects exhibited a tendency to decrease in size with time.
Premraj, Shepperd, Kitchenham, Forselius Empirical Analysis of Software Productivity
24. Presented at Metrics 2005 - Como, Italy
Past and Present Comparison
FinnishMF Finnish602 - MF
Start dates 1978-94 1997-2003
No. of companies 26 17
No. of projects 206 401
Project sizes (FPs) 33−3375 27−5060
Productivity (FPhr−1 ) 0.177 0.233
Table: Na¨ Productivity Comparison of 1978-94 and 1997-2003
ıve
Why Na¨
ıve?
Many differences between both samples of data.
Non-constant distribution of projects across business sectors.
Maintenance projects were added only 1997 onwards.
Projects exhibited a tendency to decrease in size with time.
Premraj, Shepperd, Kitchenham, Forselius Empirical Analysis of Software Productivity
25. Presented at Metrics 2005 - Como, Italy
Past and Present Comparison
FinnishMF Finnish602 - MF
Start dates 1978-94 1997-2003
No. of companies 26 17
No. of projects 206 401
Project sizes (FPs) 33−3375 27−5060
Productivity (FPhr−1 ) 0.177 0.233
Table: Na¨ Productivity Comparison of 1978-94 and 1997-2003
ıve
Why Na¨
ıve?
Many differences between both samples of data.
Non-constant distribution of projects across business sectors.
Maintenance projects were added only 1997 onwards.
Projects exhibited a tendency to decrease in size with time.
Premraj, Shepperd, Kitchenham, Forselius Empirical Analysis of Software Productivity
26. Presented at Metrics 2005 - Como, Italy
Past and Present Comparison
FinnishMF Finnish602 - MF
Start dates 1978-94 1997-2003
No. of companies 26 17
No. of projects 206 401
Project sizes (FPs) 33−3375 27−5060
Productivity (FPhr−1 ) 0.177 0.233
Table: Na¨ Productivity Comparison of 1978-94 and 1997-2003
ıve
Why Na¨
ıve?
Many differences between both samples of data.
Non-constant distribution of projects across business sectors.
Maintenance projects were added only 1997 onwards.
Projects exhibited a tendency to decrease in size with time.
Premraj, Shepperd, Kitchenham, Forselius Empirical Analysis of Software Productivity
27. Presented at Metrics 2005 - Como, Italy
Past and Present Comparison
FinnishMF Finnish602 - MF
Start dates 1978-94 1997-2003
No. of companies 26 17
No. of projects 206 401
Project sizes (FPs) 33−3375 27−5060
Productivity (FPhr−1 ) 0.177 0.233
Table: Na¨ Productivity Comparison of 1978-94 and 1997-2003
ıve
Why Na¨
ıve?
Many differences between both samples of data.
Non-constant distribution of projects across business sectors.
Maintenance projects were added only 1997 onwards.
Projects exhibited a tendency to decrease in size with time.
Premraj, Shepperd, Kitchenham, Forselius Empirical Analysis of Software Productivity
28. Presented at Metrics 2005 - Como, Italy
Productivity Model
Regression Model of the form:
ln(Effort) = βyr ln(Sizeyr ) + BusSectBool + ProjTypeBool
1 Year*Size Interaction: Each year - 1978, ..., 2003 became the
dummy variable and ln(Size) the project size in FP for the
project.
2 Boolean dummy variables for business sector.
3 Boolean dummy variables for project type (i.e. New Devp. or
Maintenance).
Premraj, Shepperd, Kitchenham, Forselius Empirical Analysis of Software Productivity
29. Presented at Metrics 2005 - Como, Italy
Productivity Model
Regression Model of the form:
ln(Effort) = βyr ln(Sizeyr ) + BusSectBool + ProjTypeBool
1 Year*Size Interaction: Each year - 1978, ..., 2003 became the
dummy variable and ln(Size) the project size in FP for the
project.
2 Boolean dummy variables for business sector.
3 Boolean dummy variables for project type (i.e. New Devp. or
Maintenance).
Premraj, Shepperd, Kitchenham, Forselius Empirical Analysis of Software Productivity
30. Presented at Metrics 2005 - Como, Italy
Productivity Model
Regression Model of the form:
ln(Effort) = βyr ln(Sizeyr ) + BusSectBool + ProjTypeBool
1 Year*Size Interaction: Each year - 1978, ..., 2003 became the
dummy variable and ln(Size) the project size in FP for the
project.
2 Boolean dummy variables for business sector.
3 Boolean dummy variables for project type (i.e. New Devp. or
Maintenance).
Premraj, Shepperd, Kitchenham, Forselius Empirical Analysis of Software Productivity
31. Presented at Metrics 2005 - Como, Italy
Productivity Model
Regression Model of the form:
ln(Effort) = βyr ln(Sizeyr ) + BusSectBool + ProjTypeBool
1 Year*Size Interaction: Each year - 1978, ..., 2003 became the
dummy variable and ln(Size) the project size in FP for the
project.
2 Boolean dummy variables for business sector.
3 Boolean dummy variables for project type (i.e. New Devp. or
Maintenance).
Premraj, Shepperd, Kitchenham, Forselius Empirical Analysis of Software Productivity
33. Presented at Metrics 2005 - Como, Italy
New Development Project Models
ANOVA highlights significant differences between project size and
effort of New Development and Maintenance projects.
Project Type Dummy Variable
βNewDevp = 0.1198
p = 0.235 and −0.091 < βNewDevp < 0.331
+ve value implies more effort for New Development projects
than Maintenance (latter being a point of reference and
hence, is zero in the dummy variable).
Results in line with Kitchenham et al - No significant
differences in productivity between New Development and
Maintenance projects.
Premraj, Shepperd, Kitchenham, Forselius Empirical Analysis of Software Productivity
34. Presented at Metrics 2005 - Como, Italy
New Development Project Models
ANOVA highlights significant differences between project size and
effort of New Development and Maintenance projects.
Project Type Dummy Variable
βNewDevp = 0.1198
p = 0.235 and −0.091 < βNewDevp < 0.331
+ve value implies more effort for New Development projects
than Maintenance (latter being a point of reference and
hence, is zero in the dummy variable).
Results in line with Kitchenham et al - No significant
differences in productivity between New Development and
Maintenance projects.
Premraj, Shepperd, Kitchenham, Forselius Empirical Analysis of Software Productivity
35. Presented at Metrics 2005 - Como, Italy
New Development Project Models
ANOVA highlights significant differences between project size and
effort of New Development and Maintenance projects.
Project Type Dummy Variable
βNewDevp = 0.1198
p = 0.235 and −0.091 < βNewDevp < 0.331
+ve value implies more effort for New Development projects
than Maintenance (latter being a point of reference and
hence, is zero in the dummy variable).
Results in line with Kitchenham et al - No significant
differences in productivity between New Development and
Maintenance projects.
Premraj, Shepperd, Kitchenham, Forselius Empirical Analysis of Software Productivity
36. Presented at Metrics 2005 - Como, Italy
New Development Project Models
7
Effort vs. Size
Power Model
6 Power Model - Outliers
Outliers
- All Projects -
Effort = 6.55 (Size)0.981
5
Effort (Hours - ∗104)
4
3
2
- Without 30 Outliers -
Effort = 5.23 (Size)1.021
1
0
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 5500
Project Size (EP20 Function Points)
Premraj, Shepperd, Kitchenham, Forselius Empirical Analysis of Software Productivity
37. Presented at Metrics 2005 - Como, Italy
New Development Project Models
7
Effort vs. Size
Power Model
6 Power Model - Outliers
Outliers
- All Projects -
Effort = 6.55 (Size)0.981
5
Effort (Hours - ∗104)
4
3
2
- Without 30 Outliers -
Effort = 5.23 (Size)1.021
1
0
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 5500
Project Size (EP20 Function Points)
Premraj, Shepperd, Kitchenham, Forselius Empirical Analysis of Software Productivity
38. Presented at Metrics 2005 - Como, Italy
New Development Project Models
7
Effort vs. Size
Power Model
6 Power Model - Outliers
Outliers
- All Projects -
Effort = 6.55 (Size)0.981
5
Effort (Hours - ∗104)
4
3
2
- Without 30 Outliers -
Effort = 5.23 (Size)1.021
1
0
0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 5500
Project Size (EP20 Function Points)
Premraj, Shepperd, Kitchenham, Forselius Empirical Analysis of Software Productivity
39. Presented at Metrics 2005 - Como, Italy
Maintenance Project Models
7
Effort vs. Size - All Projects -
Effort = 20.6 (Size)0.734
Power Model
6
Power Model - Outliers
Outliers
0.613 < b < 0.856
5
Effort (Hours - ∗103)
4
3
2 - Without 4 Outliers -
Effort = 23.5 (Size)0.718
1
0.615 < b < 0.821
0
0 100 200 300 400 500 600 700 800 900 1000
Project Size (EP20 Function Points)
Premraj, Shepperd, Kitchenham, Forselius Empirical Analysis of Software Productivity
40. Presented at Metrics 2005 - Como, Italy
Maintenance Project Models
7
Effort vs. Size - All Projects -
Effort = 20.6 (Size)0.734
Power Model
6
Power Model - Outliers
Outliers
0.613 < b < 0.856
5
Effort (Hours - ∗103)
4
3
2 - Without 4 Outliers -
Effort = 23.5 (Size)0.718
1
0.615 < b < 0.821
0
0 100 200 300 400 500 600 700 800 900 1000
Project Size (EP20 Function Points)
Premraj, Shepperd, Kitchenham, Forselius Empirical Analysis of Software Productivity
41. Presented at Metrics 2005 - Como, Italy
Maintenance Project Models
7
Effort vs. Size - All Projects -
Effort = 20.6 (Size)0.734
Power Model
6
Power Model - Outliers
Outliers
0.613 < b < 0.856
5
Effort (Hours - ∗103)
4
3
2 - Without 4 Outliers -
Effort = 23.5 (Size)0.718
1
0.615 < b < 0.821
0
0 100 200 300 400 500 600 700 800 900 1000
Project Size (EP20 Function Points)
Premraj, Shepperd, Kitchenham, Forselius Empirical Analysis of Software Productivity
42. Presented at Metrics 2005 - Como, Italy
Sources of Variance
Table: ANOVA of Productivity Factors
Variable % of variance “explained”
Company 26.2
Process model 12.6
Business sector 11.7
Year 8.4
Hardware 5.6
ANOVA performed on Factors against productivity.
Variables significant at p = 0.01.
Premraj, Shepperd, Kitchenham, Forselius Empirical Analysis of Software Productivity
43. Presented at Metrics 2005 - Como, Italy
Productivity across Companies
Finnish602 comprises 32 companies.
Removing infrequent companies (5 or less projects) reduces
variance explained to 21.1%.
Results in line with analysis by Maxwell and Forselius.
Is Company acting as a proxy for Business Sector?
Cross-tabulating both factors shows companies almost exclusively
develop projects within a single business sector.
Choice of many factors (technical and non-technical) are determined
by business sectors e.g. staff skills, process models, security
requirements, etc.
Premraj, Shepperd, Kitchenham, Forselius Empirical Analysis of Software Productivity
44. Presented at Metrics 2005 - Como, Italy
Productivity across Companies
Finnish602 comprises 32 companies.
Removing infrequent companies (5 or less projects) reduces
variance explained to 21.1%.
Results in line with analysis by Maxwell and Forselius.
Is Company acting as a proxy for Business Sector?
Cross-tabulating both factors shows companies almost exclusively
develop projects within a single business sector.
Choice of many factors (technical and non-technical) are determined
by business sectors e.g. staff skills, process models, security
requirements, etc.
Premraj, Shepperd, Kitchenham, Forselius Empirical Analysis of Software Productivity
45. Presented at Metrics 2005 - Como, Italy
Productivity across Companies
Finnish602 comprises 32 companies.
Removing infrequent companies (5 or less projects) reduces
variance explained to 21.1%.
Results in line with analysis by Maxwell and Forselius.
Is Company acting as a proxy for Business Sector?
Cross-tabulating both factors shows companies almost exclusively
develop projects within a single business sector.
Choice of many factors (technical and non-technical) are determined
by business sectors e.g. staff skills, process models, security
requirements, etc.
Premraj, Shepperd, Kitchenham, Forselius Empirical Analysis of Software Productivity
46. Presented at Metrics 2005 - Como, Italy
Productivity across Companies
Finnish602 comprises 32 companies.
Removing infrequent companies (5 or less projects) reduces
variance explained to 21.1%.
Results in line with analysis by Maxwell and Forselius.
Is Company acting as a proxy for Business Sector?
Cross-tabulating both factors shows companies almost exclusively
develop projects within a single business sector.
Choice of many factors (technical and non-technical) are determined
by business sectors e.g. staff skills, process models, security
requirements, etc.
Premraj, Shepperd, Kitchenham, Forselius Empirical Analysis of Software Productivity
47. Presented at Metrics 2005 - Como, Italy
Productivity across Companies
Finnish602 comprises 32 companies.
Removing infrequent companies (5 or less projects) reduces
variance explained to 21.1%.
Results in line with analysis by Maxwell and Forselius.
Is Company acting as a proxy for Business Sector?
Cross-tabulating both factors shows companies almost exclusively
develop projects within a single business sector.
Choice of many factors (technical and non-technical) are determined
by business sectors e.g. staff skills, process models, security
requirements, etc.
Premraj, Shepperd, Kitchenham, Forselius Empirical Analysis of Software Productivity
48. Presented at Metrics 2005 - Como, Italy
Productivity across Companies
Finnish602 comprises 32 companies.
Removing infrequent companies (5 or less projects) reduces
variance explained to 21.1%.
Results in line with analysis by Maxwell and Forselius.
Is Company acting as a proxy for Business Sector?
Cross-tabulating both factors shows companies almost exclusively
develop projects within a single business sector.
Choice of many factors (technical and non-technical) are determined
by business sectors e.g. staff skills, process models, security
requirements, etc.
Premraj, Shepperd, Kitchenham, Forselius Empirical Analysis of Software Productivity
49. Presented at Metrics 2005 - Como, Italy
Productivity across Companies
Finnish602 comprises 32 companies.
Removing infrequent companies (5 or less projects) reduces
variance explained to 21.1%.
Results in line with analysis by Maxwell and Forselius.
Is Company acting as a proxy for Business Sector?
Cross-tabulating both factors shows companies almost exclusively
develop projects within a single business sector.
Choice of many factors (technical and non-technical) are determined
by business sectors e.g. staff skills, process models, security
requirements, etc.
Premraj, Shepperd, Kitchenham, Forselius Empirical Analysis of Software Productivity
50. Business Sector Productivity Comparison
0.336
Manufacturing
0.337
0.279
Retail
0.253
0.270
Public Admin.
0.232
0.237
Banking
0.116
Pre - 1995 Projects
Post - 1996 Projects
0.191
Insurance
0.116
0.240
Other
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35
52. Presented at Metrics 2005 - Como, Italy
Conclusions
Analysis
Overall increase of 33% in productivity.
Strongest increase in productivity during 1980s and early
1990s.
No evidence of diseconomies of scale, but pronounced
evidence of economies of scale for Maintenance projects.
Little difference between productivity of New Development
and Maintenance projects.
Most significant factors - Company, Business Sector, Year and
Hardware.
Problem of generalisation.
Premraj, Shepperd, Kitchenham, Forselius Empirical Analysis of Software Productivity
53. Presented at Metrics 2005 - Como, Italy
Conclusions
Analysis
Overall increase of 33% in productivity.
Strongest increase in productivity during 1980s and early
1990s.
No evidence of diseconomies of scale, but pronounced
evidence of economies of scale for Maintenance projects.
Little difference between productivity of New Development
and Maintenance projects.
Most significant factors - Company, Business Sector, Year and
Hardware.
Problem of generalisation.
Premraj, Shepperd, Kitchenham, Forselius Empirical Analysis of Software Productivity
54. Presented at Metrics 2005 - Como, Italy
Conclusions
Analysis
Overall increase of 33% in productivity.
Strongest increase in productivity during 1980s and early
1990s.
No evidence of diseconomies of scale, but pronounced
evidence of economies of scale for Maintenance projects.
Little difference between productivity of New Development
and Maintenance projects.
Most significant factors - Company, Business Sector, Year and
Hardware.
Problem of generalisation.
Premraj, Shepperd, Kitchenham, Forselius Empirical Analysis of Software Productivity
55. Presented at Metrics 2005 - Como, Italy
Conclusions
Analysis
Overall increase of 33% in productivity.
Strongest increase in productivity during 1980s and early
1990s.
No evidence of diseconomies of scale, but pronounced
evidence of economies of scale for Maintenance projects.
Little difference between productivity of New Development
and Maintenance projects.
Most significant factors - Company, Business Sector, Year and
Hardware.
Problem of generalisation.
Premraj, Shepperd, Kitchenham, Forselius Empirical Analysis of Software Productivity
56. Presented at Metrics 2005 - Como, Italy
Conclusions
Analysis
Overall increase of 33% in productivity.
Strongest increase in productivity during 1980s and early
1990s.
No evidence of diseconomies of scale, but pronounced
evidence of economies of scale for Maintenance projects.
Little difference between productivity of New Development
and Maintenance projects.
Most significant factors - Company, Business Sector, Year and
Hardware.
Problem of generalisation.
Premraj, Shepperd, Kitchenham, Forselius Empirical Analysis of Software Productivity
57. Presented at Metrics 2005 - Como, Italy
Conclusions
Analysis
Overall increase of 33% in productivity.
Strongest increase in productivity during 1980s and early
1990s.
No evidence of diseconomies of scale, but pronounced
evidence of economies of scale for Maintenance projects.
Little difference between productivity of New Development
and Maintenance projects.
Most significant factors - Company, Business Sector, Year and
Hardware.
Problem of generalisation.
Premraj, Shepperd, Kitchenham, Forselius Empirical Analysis of Software Productivity
58. Presented at Metrics 2005 - Como, Italy
Conclusions
Analysis
Overall increase of 33% in productivity.
Strongest increase in productivity during 1980s and early
1990s.
No evidence of diseconomies of scale, but pronounced
evidence of economies of scale for Maintenance projects.
Little difference between productivity of New Development
and Maintenance projects.
Most significant factors - Company, Business Sector, Year and
Hardware.
Problem of generalisation.
Premraj, Shepperd, Kitchenham, Forselius Empirical Analysis of Software Productivity
59. Presented at Metrics 2005 - Como, Italy
Conclusions
Process
Large data sets are hard to analyse and it is easy to
misunderstand the data.
Encourage contact with the data collecting entity.
This is an initial analysis that has scratched the surface of a
large data set.
Premraj, Shepperd, Kitchenham, Forselius Empirical Analysis of Software Productivity
60. Presented at Metrics 2005 - Como, Italy
Conclusions
Process
Large data sets are hard to analyse and it is easy to
misunderstand the data.
Encourage contact with the data collecting entity.
This is an initial analysis that has scratched the surface of a
large data set.
Premraj, Shepperd, Kitchenham, Forselius Empirical Analysis of Software Productivity
61. Presented at Metrics 2005 - Como, Italy
Conclusions
Process
Large data sets are hard to analyse and it is easy to
misunderstand the data.
Encourage contact with the data collecting entity.
This is an initial analysis that has scratched the surface of a
large data set.
Premraj, Shepperd, Kitchenham, Forselius Empirical Analysis of Software Productivity
62. Presented at Metrics 2005 - Como, Italy
Conclusions
Process
Large data sets are hard to analyse and it is easy to
misunderstand the data.
Encourage contact with the data collecting entity.
This is an initial analysis that has scratched the surface of a
large data set.
Premraj, Shepperd, Kitchenham, Forselius Empirical Analysis of Software Productivity
63. Presented at Metrics 2005 - Como, Italy
End of Presentation
Authors –
1 Rahul Premraj – rpremraj@bmth.ac.uk
2 Martin Shepperd – martin.shepperd@brunel.ac.uk
3 Barbara Kitchenham – barbara.kitchenham@nicta.com.au
4 Pekka Forselius – pekka.forselius@kolumbus.fi
Thank you for your attention.
Questions please!
Premraj, Shepperd, Kitchenham, Forselius Empirical Analysis of Software Productivity