Data form a set of primary studies on reading methods for software inspections. They were analysed by Lech Madeyski and Barbara Kitchenham, 'How variations in experimental designs impact the construction of comparable effect sizes for meta-analysis', 2015.
MadeyskiKitchenham.MetaAnalysis.PBRvsCBRorARA data frame with 17 rows and 26 variables:
Name of empirical study
Reference to the paper reporting primary study or experimental run where data were originally reported
The number of teams including both, PBR and Control teams
Experimental design description: Before-after, Between-groups, Cross-over
Experimental design: between-groups (BG), within-subjects cross-over (WSCO), within-subjects before-after (WSBA)
The average proportion of defects found by teams using PBR
The average proportion of defects found by teams using Control treatment: Check-Based Reading (CBR) or Ad-Hoc Reading (AR)
The difference between M_PBR and M_C, i.e. Diff = M_PBR - M_C
The percentage increase in defect rate detection, i.e. Inc=100*[(M_PBR-M_C)/M_C]
The standard deviation of the control group values reported by the original Authors, i.e., obtained from the papers/raw data
The standard deviation of the control group values equals SD_C_ByAuthors for studies for which the data was available OR the weighted average of SD_C_ByAuthors (i.e., 0.169) for studies where SD_C_ByAuthors is missing.
The variance of the Control group observations, i.e., the variance obtained from the teams using the Control method V_C=SD_C^2
The variance of the unstandardized mean difference D (between the mean value for the treatment group and the mean value for the Control group)
This is the equivalent of SD_C (the standard deviation of the control group) based on a different variance for the student studies or the practitioner studies depending on the subject type of the study with the missing value.
The variance of the mean difference in the meta-analysis based on SD_C_Alt
The sum of squares of the Control group values. For within subjects studies SS=V_C*(n-1). For between subjects studies SS=V_C*(n_C-1)
The number of PBR teams
The number of Control (CBR or AR) teams
Type of Control treatment: CRB or AR
Type of participants: Engineers or Students
Type of team: Nominal or Real
Reflects size of the teams: 2-PersonTeam or LargerTeam
The type of artefact: Requirements or Other
Whether study is associated with Basili (the forerunner): Yes or No
Combined ControlType and AssociatedWithBasili: AH_AssociatedWithBasili, CBR_AssociatedWithBasili, CBR_NotAssociatedWithBasili
If you use this data set please cite: Lech Madeyski and Barbara Kitchenham, 'How variations in experimental designs impact the construction of comparable effect sizes for meta-analysis', 2015.
MadeyskiKitchenham.MetaAnalysis.PBRvsCBRorAR
Run the code above in your browser using DataLab