E-Health (EHL)

1
Summary of day 1

• Introduction to Knowledge Mapping and Social
  Network Analysis (SNA)
• The importance of asking the right question

Questions ?
2
Day 2
3
Session 4-Data collection
Steeve Ebener, WHO
Manila, Philippines 28-30 April 2008
4
Content

1. Which data to collect
2. Source of the data
3. Ego vs Full network
4. Ego network data collection
5. Full network data collection
6. Data collection Instrument
7. Data issues
8. Example of the DTTB project

5
Which data to collect ?

• Several types
• data about the respondent (demographics)
• network specific questions (see previous session)
• any additional information which might be useful
  for the analysis
• case of the Ego networks
• example of the DTTB project

6
Which data to collect ?
Data about the respondent (demographics)
Examples

• income
• education
• location (birth, assignment,...)
• gender
• age
• ethnicity
• religion
• occupation
• ...

Valente
7
Which data to collect ?
Other data of importance for the analysis
Examples

• DTTB project
• Access to internet
• Area of expertise they have been looking for
• Number of experts they knew before being deployed
• Electricity availability
• Name of the referral facility
• What to improve in the DTTB program
• time and reason of contact

• Cultural values
• Competency areas
• Use of different types of media (online, printed,
  face-to-face) for getting information.

Experts directory
8
Source of the data

• Secondary data (often 2 mode)
• Memberships in groups
• Facebook "networks"
• Participation in events
• Listserv threads
• Voting records
• Text analysis
• Other
• Email records, purchase/sale records, marriage
  records, etc... (patient records?)

Borgatti
9
Source of the data

• Primary data
• Experiments
• Rumor planting
• Observation
• Westen-Electric Hawthorne plant studies
• Survey/Census/Logbook
• Telephone, web, paper, etc
• example of the DTTB project

Borgatti
10
Ego vs full network

• Ego Network data collection
• interview of the respondents (egos) to ask about
  their contacts (alters)
• The alters are not interviewed
• One ego's alters are not matched up with other
  egos or their alters
• Collect a lot of (perceived) info on the alters
• Full Network data collection ("regular" SNA)
• interview of the respondents and their contacts
• Do generally not collect info on the alters

Borgatti
11
Ego vs full network
Borgatti
12
Ego Network data collection

• Characterize the relationship with each alter
• Optionally obtain ego's perception of which
  alters have ties with other alters
• Connections between ego's or between alters of
  different egos are not collected

Modified from Borgatti
13
Full Network data collection

• Survey vs Census
• in a survey only part of the network is
  interviewed
• sampling issue (still in its infancy)
• data analyzed but not graphed
• in a census all the network is interviewed
• saturation sampling
• The most typical kind of network methodology
• Usually what people think of when we say
  networks.
• These data can be graphed and analyzed using
  matrix methods

Modified from Borgatti and Valente
14
Full Network data collection

• Survey - Sampling
• Fixed probability (e.g. random sampling)
• Fine with Ego Network
• Ok for some complete network studies
• Adaptive sample (e.g. snowball sampling)

Red nodes are interviewed alters Blue node are
not Interviewed

• Stances
• Nominalist/etic (least delusional approach)
• Realist/emic (best used for true groups)
• Combination

Modified from Borgatti and Valente
15
Full Network data collection
Survey sampling (Stance)
Borgatti
16
Full Network data collection

• Census
• Unknown network
• Ask for the help of one or several of the members
• Use the snowball approach to identify all the
  members
• Known network
• Interview all the members

Avoid survey when possible !
Borgatti and Valente
17
Data collection instruments
Ethical Issues

• Respondents cannot be anonymous
• Snowball sampling ask respondent to name others
• "bill says that you inject illegal drugs with
  him, can we talk to you ?
• Missing data are troublesome
• Might put the attention on the wrong issue
• Results may be wrong
• Non-participants still included as mentioned by
  others
• And participants are like informers
• Outputs ideally show individual level data
• Pushes boundary of the professional
• Deceptively powerful
• SNA is still unknown, look like research
• Quid pro quo arrangements with research sites
• Management might hire/fire based on results

Borgatti
18
Data collection instruments
Ethical Issues
Need to find out what are the potential ethical
threat and to whom ?

• In academic setting
• In management setting
• In mixed situations
• In national security setting
• ...

Need to address them
Borgatti
19
Data collection instruments
Ethical Issues

• Consent form, disclosure contract
• Anonymizing (not releasing demographic data for
  example)
• Address non-participation
• Aggregating categorizing
• Avoid Quid pro quos
• Find the good time to perform the analysis (not
  before a restructuration for example)
• Protect data (e.g. theft)
• Organization consulting (who gets to see the data
  ?, professional debriefing,...)
• ...

Borgatti
20
Data collection instruments
Ethical issues
Truly informed consent form
Borgatti
21
Data collection instruments
Ethical issues
3-way disclosure contract

• For research done in organizations

• Signed by management, the researchers, and each
  participants

• Clearly identifies what will be done with the data

Borgatti
22
Data collection instruments
Ethical issues
Questionnaire might also have to go through the
ethical committee's approval
Confidentiality reminder on the questionnaire
Borgatti
23
Data collection instruments
Which instrument ?

• Data about the respondent (demographics)
• Network specific questions
• Any additional information which might be useful
  for the analysis

Questionnaire form(s) if performed once
Wants to do it over time ?
Logbook
DTTB project
24
Data collection instruments
Questionnaire format
General questionnaire rules also applies here
(e.g. importance of the order of the questions)
Some specific issues

• Aided (rosters) vs unaided (open-ends)
• Tick, rate or rank ?
• Across (Multigrids) or down (separated questions)
• Paper or electronic

Borgatti
25
Data collection instruments
Closed-Ended vs Open-Ended Roster of names or
just blank lines ?

• Closed-ended (aided)
• requires bounded list
• Can be impractical for large network
• Each alter has equal chance of choice
• Open-ended (unaided)
• Subject to recall errors
• can limit number of choices made
• (more effort, limited space)

use hybrid designs otherwise
Borgatti
26
Data collection instruments
Hybrid Questionnaire

• Paper questionnaire with a separated booklet
  containing name directory
• Web version questionnaire with drop-down menus

Hybrid designs are useful in large networks
Importance to use a unique ID to cover name
writing mistakes
Borgatti
27
Data collection instruments
Tick, rate or rank ?

• Ask respondent for yes/no decisions or
  quantitative assessment ?
• Yes/no are easier on respondents (therefore
  reliable, believable
• Yes/no "much" faster to administer
• But yes/no provides no discrimination among
  levels ratings provide more nuance
• A series of binaries can replace on quantitative
  rating
• Instead of "How-often do you see each person?"
• 1 once a year 2 once a month 3 once a week
  etc.
• Use three questions (in this order)
• Who do you see at least once a year ?
• Who do you see at least once a month ?
• Who do you see at least once a week ?

Borgatti
28
Data collection instruments
Tick, rate or rank ?

• Users asked to rank others in terms of
  communication frequency over 3 weeks
• person most communicated with was ranked in top 4
  only 52 of time
• Accuracy of prediction s of the next month's
  communication was the same, who you like who
  you talk to
• All studies taken together had similar results
• Studies in other fields corroborate

Ranking can give unreliable results !
Borgatti
29
Data collection instruments
Repeated Roster vs MultiGrid
Borgatti
30
Data collection instruments
Paper or electronic ?

• Paper medium
• Reliable
• Reassuring to respondents
• Possible errors in data entry
• Data entry is time-consuming
• ...
• Electronic (Pda, Web)
• Span distances, time zones
• harder to lose
• Fewer data handling errors
• Possible lower response rate
• Emailed documents (e.g. excel file) vs online
  survey instrument
• ...

Choose what is most adapted
Borgatti
31
Data collection instruments
Design Considerations

• Network questionnaires can be fun but are usually
  time-consuming and might generate anxiety
• Providing value
• Treating respondents with respect
• Attractive formatting
• Cloak in authority and importance
• Do not forget that multiple, similar relational
  questions risk respondent fatigue and annoyance
• Who do you give advice to ?
• Who do you give information to ?
• Who do you give guidance to ?
• Who do you counsel ?

Aggregation to larger categories, such as
affective instrumental can work well
Borgatti
32
Data issues
Information accuracy

• Response strategies and biases
• Appearing central
• Recalling important people more than others
• Hiding illicit relationships
• Cognitive cultural schemas
• Recall biased towards normal, frequent, logical
• Role schemas filter perceptions, learning

Can't do much about it but need to be aware of it
Ethnographic sandwich
Borgatti
33
Data issues
Unexpected asymmetry

• A claims to have sex with B, but B does not claim
  to have sex with A
• The relation is logically symmetric, but
  empirically asymmetric
• can be an error in the recall, strategic response
• Sometimes asymmetry is the point
• Logically symmetric data may be symmetrized
• if either A or B mention the other, it's a tie

• ! Can't symmetrize logically non-symmetric
  relations (e.g. gives advices to), unless
• changing the meaning of tie
• you have asked the question both ways (who do you
  give advice to ?, who gives advice to you ?)

Borgatti
34
Data issues
Missing data

• Quick and dirty
• For logically symmetric relations
• if B-A tie is missing, substitute A-B tie
• For logically non-symmetric relations, ask
  questions both ways (see previous slide)
• Bayesian imputation methods (not addressed here)

Borgatti
35
Example of the DTTB Project

• Ego or Full network data collection ?
• Known network
• 2 Batches (18 in Batch 21 and 16 in Batch 22)
• A DTTB doctors can contact and other DTTB doctor
  or an outside expert. Further contacts made by
  outside experts are not captured
• mixed Ego (contact with experts) and full network
  (among DTTBs) analysis
• use of expert information form
• Question who do you contact for medical
  expertise ?
• Expertise network analysis
• Data collected over a period of 9 months
• time dimension

Mixed approach
36
Example of the DTTB Project
Questionnaires and forms
Questionnaire Batch 21- 22
Questionnaire previous Batches
Experts contact form
37
Example of the DTTB Project
Logbook