2/17/2016
HEALTH RESEARCH FUNDAMENTALS
NIeCer 101
National Institute of Epidemiology
Chennai
[Link]
Data management includes
HEALTH RESEARCH FUNDAMENTALS
Define variables
NIeCer 101
Create study database and data dictionary
National Institute of Epidemiology
Chennai
Enter data and correct errors
Create dataset for analysis
Back up and archive the dataset
[Link]
1
2/17/2016
Key elements of data management
HEALTH RESEARCH FUNDAMENTALS
• Data structure
NIeCer 101
• Data entry
• Individual and aggregated databases
National Institute of Epidemiology
Chennai
• Mother and daughter databases
[Link]
Basic structure of a database
HEALTH RESEARCH FUNDAMENTALS
• Lines represent records
• Columns represent variables
NIeCer 101
Identifier Variable 1 Variable 2 Variable 3 Variable 4 Etc…
National Institute of Epidemiology
Chennai
Record 1
Record 2
Record 3
Etc…
[Link]
Structure
2
2/17/2016
Data documentation
HEALTH RESEARCH FUNDAMENTALS
• Structure
• Name, number of records etc
NIeCer 101
• Variables
• Name, values, coding
National Institute of Epidemiology
Chennai
• History
• Creation, modification
• Storage information
• Media, location, back up
• Additional information
[Link]
Structure
Identifier in the database
HEALTH RESEARCH FUNDAMENTALS
• Unique
NIeCer 101
• Maintained by a computerized index
• Secured by quality assurance procedures
National Institute of Epidemiology
Chennai
[Link]
Structure
3
2/17/2016
Using codes within the unique identifier
HEALTH RESEARCH FUNDAMENTALS
Village
• Unique identifier may
contain all information about Street
NIeCer 101
that particular ID House
National Institute of Epidemiology
Chennai
• Each digit or set of digits refer Person
to specific information 1 2 3 4 5 6 7
• Example:
• First and second digit: village
• Third and fourth digit: Street
• Fifth digit: House
• Sixth and seventh digit: Person [Link]
Structure
Structure of the variables in the database
HEALTH RESEARCH FUNDAMENTALS
• Integer
• Specify the number of digits NIeCer 101
• Numeric
National Institute of Epidemiology
Chennai
• Specify the number of decimals
• Alpha-numeric
• Specify length
• Turn all letters to capitals
• Dates (specific format)
[Link]
Structure
4
2/17/2016
Creating variable names
HEALTH RESEARCH FUNDAMENTALS
• Clear
• Need to refer to the questionnaire item
• Understandable (e.g., “EXERDAILY” for “Exercise daily”)
NIeCer 101
• Short, no space
• Most softwares require less than 10 characters
National Institute of Epidemiology
Chennai
• Consistent
• “EXERPAST” for “Exercise daily in the past”
• “EXERCURRDLY” for “Exercise daily in the current ”
• “EXERPASTOCC” for “Exercise occasionally in the past”
• “EXERCURROCC” for “Exercise occasionally in the current”
• “VARIAB” for all crude variables (EXERCISE)
• “VARIAB_12” for all dichotomized variables (EXERCISE_12)
• No duplicate
• Trimming of names by software can create duplicate name Structure
[Link]
Design data entry-friendly data collection
HEALTH RESEARCH FUNDAMENTALS
instrument
• Outline
• Identifiers NIeCer 101
• Demographics
National Institute of Epidemiology
Chennai
• Outcome (Health problem/disease)
• Exposures (variables, including third factors)
• Auto-coding function
[Link]
Entry
5
2/17/2016
Coding
HEALTH RESEARCH FUNDAMENTALS
• Prefer numerical coding
• Decide on
NIeCer 101
• Missing values (.) or (9, 99, 999)
• Not applicable (8, 98, 998)
National Institute of Epidemiology
Chennai
• Avoid cumbersome codes
• WALKING (1) and CYCLING (2)
• Doing WALKING and CYCLING (12)
• Use as “1” or “0” (“1” or “2”) as baseline for
gradients (Yes/No or Present/Absent) as appropriate
depending on software for analysis [Link]
Entry
Constructing a data dictionary
HEALTH RESEARCH FUNDAMENTALS
• Contains, for each variable:
Question Variable Type Format Values Logical
• Variable name name checks
NIeCer 101
• Description of questionnaire 1 EXERDAILY Integer Yes =1 Skip
No =2 pattern
item
2 EXERTYPE Integer Walking =1
National Institute of Epidemiology
Chennai
• Various values of variable Cycling =2
(e.g., 1, 2, 3)
ETC…
• Meaning of each value (e.g.,
1= Yes, 2=No) Some softwares create variable catalogue automatically; Ideally investigator constructs the same
• The catalogue is particularly useful:
• When a database is shared with others
• If the researcher has to get back to the database later [Link]
Entry
6
2/17/2016
Check specifications before data entry
HEALTH RESEARCH FUNDAMENTALS
• Minimum and maximum values
NIeCer 101
• Legal codes
• Set of values that will be accepted
National Institute of Epidemiology
Chennai
e.g., 1, 0 and 9 for “Yes”, “No” and “Missing”
• Skip patterns
• Automatic coding
• Copying data from preceding record
• Calculations [Link]
Entry
Data entry
HEALTH RESEARCH FUNDAMENTALS
• Use as opportunity for partial data cleaning
• Write comments NIeCer 101
• Seek clarification
National Institute of Epidemiology
Chennai
• Use checks
• Mark each paper as data entry is completed
• Validate after data entry
[Link]
Entry
7
2/17/2016
Individual and aggregated databases
HEALTH RESEARCH FUNDAMENTALS
• Individual databases
NIeCer 101
• Each record is an observation
• Aggregated database
National Institute of Epidemiology
Chennai
• Records contain counts
• Normalized database
• Only one count by record
• Facilitates further aggregation
[Link]
Individual and aggregated databases
HEALTH RESEARCH FUNDAMENTALS
Aggregating individual data
Individual data Aggregated file NIeCer 101
I Place Age Sex Onset
D I Place Count
National Institute of Epidemiology
Chennai
D
1 A 3 1 1 Jan 06
1 A 5
2 B 1 2 1 Jan 06
2 B 3
3 C 35 2 3 Jan 06
3 C 37
4 D 67 1 4 Jan 06
4 D 67
5 A 2 1 2 Jan 06
6 B 2 1 4 Jan 06
5 C 2 1 5 Jan 06
… … … … … [Link]
Individual and aggregated databases
8
2/17/2016
Mother and daughter databases
HEALTH RESEARCH FUNDAMENTALS
• Information is available at various levels
Village
NIeCer 101
•
• Household
National Institute of Epidemiology
Chennai
• Individual
• Illness episode
• Store information at each level in separate
databases
• Link databases together with identifiers
[Link]
Mother and daughter databases
Mother and daughter databases
HEALTH RESEARCH FUNDAMENTALS
Household level data Individual level data
HousI Location Communit HousInco HousID PersonID Diseased Exposed
D y m 1 101 1 1
NIeCer 101
1 A 3 1 1 102 2 1
2 B 1 2 2 201 2 2
National Institute of Epidemiology
Chennai
3 C 35 2 2 202 1 2
4 D 67 1
5 E 2 1 • Each database has its own
6 F 2 1 unique identifier
5 G 2 1
• Link these relational databases
… … … …
using a common index identifier
• Merge files when needed [Link]
Mother and daughter databases
9
2/17/2016
Summing up on data management
HEALTH RESEARCH FUNDAMENTALS
• Code database numerically
NIeCer 101
• Enter data using quality assurance procedures
• Store information at the level where it needs to
National Institute of Epidemiology
Chennai
be stored
• Relate/Merge files when needed and as required
[Link]
HEALTH RESEARCH FUNDAMENTALS
NIeCer 101
National Institute of Epidemiology
Chennai
[Link]
10