Page 1
© 2006 Van Alstyne, Brynjolfsson & Aral
Information, Social Networks& Individual Success
MIT Center for E-Business / Boston UniversityMarshall Van Alstyne
With S. Aral, E. Brynjolfsson, N. Bulkley, N. Gandal, C. King, J. ZhangSponsored by NSF #9876233, Intel Corp & BT
[email protected]
© 2006 All Rights Reserved
Page 2
© 2006 Van Alstyne, Brynjolfsson & Aral
Page 3
© 2006 Van Alstyne, Brynjolfsson & Aral
IT and Productivity: The Data Speak
IT Stock (relative to industry average)
Productivity(relative to industry average)
Computers are associated with greater productivity...
...But what explains the substantial variation across firms?
Page 4
Agenda
• Study overview & technology• Visualizing organizational information and
social networks.• Participant perceptions (surveys)• Statistical models of behavior and output• Notable correlations
Page 5
© 2006 Van Alstyne, Brynjolfsson & Aral
The Current Study• Three firms initially• Unusually measurable inputs and outputs
– 1300 projects over 5 yrs and – 125,000 email messages over 10 months (avg 20% of time!)– Metrics
(i) Revenues per person and per project, (ii) number of completed projects, (iii) duration of projects, (iv) number of simultaneous projects, (v) compensation per person
• Main firm 71 people in executive search (+2 firms partial data)– 27 Partners, 29 Consultants, 13 Research, 2 IT staff
• Four Data Sets per firm – 52 Question Survey (86% response rate)– E-Mail– Accounting– 15 Semi-structured interviews
Page 6
© 2006 Van Alstyne, Brynjolfsson & Aral
The Setting – Executive Recruiting
Executive Search Process1. Partner brings in client contract2. Partner negotiates internal labor market to compose a team with
consultants and researchers (load balancing and regional approval)3. A Phased Search (Matching) Process with information inputs / outputs :
CaptureRequirements
Initial Search /Create Initial
Pool
Vet CandidatesConduct Due
Diligence
Create InterviewPool / Interview
Internally
Create FinalPool / Facilitate
Client Placement(~ 6)
Firm uses IT in 2 ways:1. Communication Vehicle (e.g. Phone, Email)2. Executive Search System (ESS) – a proprietary KMS
Internal Task Coordination (e.g. Assign Tasks & Labor ) External Contract Coordination (e.g. anti-poaching provisions) Knowledge Search (e.g. Candidates, Clients) including external DBs
Page 7
© 2006 Van Alstyne, Brynjolfsson & Aral
Tools & Technology
Organizations under an E-Mail Microscope
Page 8
© 2006 Van Alstyne, Brynjolfsson & Aral
Gaining access to live e-mail
To: Marshall Van Alstyne <[email protected] > Subject: Re: YOUR PROPOSAL Date: Sun, 17 Nov 2002 09:54:23 -0500 Cc: [email protected] , Geoffrey Parker <[email protected] > X-Originating-IP: 68.41.189.43
Ok, i will look for all the pieces today then and try to get everything in Fastlane tonight.
Meeting is up to you. I have to go to DRDA first thing in the morning to hand them all the PAFs so they can process all the proposals. The meeting is to give you one last chance to view the entire proposal package before DRDA pushes the "Send" button. We could also try to do this virtually so neither of us has to travel to the other site.
As far as footers go, let's not worry about it as long as you are page numbering each section individually. I usually add more information to the footer but I don't have time to worry about this detail.
Ann
Stop words are dropped; then the raw text is root-stemmed (e.g. “are”->“is”, “pieces”->“piece”), counted, and hashed.
Page 9
© 2006 Van Alstyne, Brynjolfsson & Aral
AnnMessage-ID: 00000000C74E9F197619354B912FA038789E97DD070095FBFC9E5C710C45AD83BE1BA97654F300000025D7D7000095FBFC9E5C710C45AD83BE1BA97654F30000015D02090000 Date: 11/17/2002 09:54:23 PM From: ChiUserWWW2 To: ChiUserWWW34 CC: ChiUserWWW2 , ChiUserEEE137 Subject: 2234380046220310381 -4543232654336644202 3187911263930032313 - 8725299062034745550 6646063218832296471 Content: -7488330257252326972<8>; 3461049762598860849<5>; -4469441121190040841<4>; 4122472038465781083<4>; - 2485003116886841409<3>; 8003219831352894262<3>; 1698764591947117759<2>; 5894537654329429962<2>; - 9076192449175488644<2>; 7750988586697557362<2>; 8871153132300476476<2>; - 7527789141644698404<2>; 8763687632651980147<1>; 3129683954660429336<1>; -6916544271211441138<1>; 6293576012604293570<1>; - 320692498224125839<1>; 8934872354483414290<1>; -6836405471713717833<1>; - 5975878511407257679<1>; -3014223241434893634<1>; - 8934856908841293615<1>; -857818984403519253<1>; 1344343662225282497<1>; 965941123633882107<1>; -3147930629716878416<1>; 7137519577624117188<1>; 7523708256417630601<1>; -6946268052250097500<1>; Attachment Number: 0 Attachment list:
This is what we “see”
Reconstructing semantics is difficult. We do not read attachments but do record type & size information (e.g. 157kb .doc file)
Page 10
© 2006 Van Alstyne, Brynjolfsson & Aral
The Survey
• 52 Questions– personal characteristics– time-use– value of tasks– technology skills – technology use– information sources– work habits– information sharing– perceptions
86% response rate
Page 11
© 2006 Van Alstyne, Brynjolfsson & Aral
Email habits show work patterns
Page 12
© 2006 Van Alstyne, Brynjolfsson & Aral
An E-mail “Fingerprint”
Consultant - Sent vs. Received
-12000
-10000
-8000
-6000
-4000
-2000
0
2000
4000
6000
8000
c10
c12
c14
c16
c18 c2 c2
1c2
3c2
7c2
9c3
0 c6 c7 c71 c9
External
Internal
Sent
Received
Page 13
Topology
Comprehending the Social Networks
Page 14
© 2006 Van Alstyne, Brynjolfsson & Aral
Clustering example from our data
Theoretically, Information Should Matter: Both Levels and Structure
Constrained
Unconstrained
Page 15
© 2006 Van Alstyne, Brynjolfsson & Aral
Social Network Efficiencies1. Connect to hubs
• Central nodes who bridge structural holes are significantly more effective.
2. Send short messages• Consultants have higher
billings (.56, p<.01) and are more central (see 1).
3. Communicate declarative information
• Gets better reply rates.• Procedural tips shared
laterally not across hierarchy (or better FTF)
4. Career Ladder• Explore early vs. exploit late
Page 16
© 2006 Van Alstyne, Brynjolfsson & Aral
Survey Summaries
Incentives & Behaviors
Page 17
© 2006 Van Alstyne, Brynjolfsson & Aral
There are culture differences. One firm shares more. Most disagree that info never enters DB
Responses to Information Sharing Questions 1-4
-1.00
-0.50
0.00
0.50
1.00
1.50
2.00
2.50
3.00
Firm X
Firm Y
Firm Z
Q1 Colleagues give me credit for info that I share.
Q3: I volunteer all relevant info to colleagues.
Q2 Colleagues willingly share their private search info with me.
Q4: A lot of my personal knowledge never reaches the corp. database.
Page 18
© 2006 Van Alstyne, Brynjolfsson & Aral
Incentive theory works
Weighting of Compensation Structure
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%
Firm X Firm Y Firm Z
Whole company performance
Project team(s) performance
Individual performance
Least Most Med.
Narrower incentives mean narrower info sharing.
Page 19
© 2006 Van Alstyne, Brynjolfsson & Aral
Firm X automates more processesPerceptions of IT Applications
-1.00-0.80-0.60-0.40-0.200.000.200.400.600.801.001.20
Firm X
Firm Y
Firm Z
Q7 We use info sys to coord sched & project handoffs
Q14 My data requirements are routine
Q15 For routine info, the process of getting it is automated
Q41 We mine our data for correlations and new ideas
Page 20
Perceived Information Overload
• Bears little correlation with e-mail received.
• Falls with increasing IT proficiency.
• Rises with colleague response delays.
• Falls with increased support staff contact.
Page 21
© 2006 Van Alstyne, Brynjolfsson & Aral
Emails “pose threat to IQ”Lack of discipline responding to email reduced productivity by the equivalent of 1 night’s sleep.
“…average IQ loss was measured at 10 points, more than double the four point mean fall found in studies of cannabis users.”
Similarly, in our study, time spent and volume processedbear little correlation with productivity…
Page 22
© 2006 Van Alstyne, Brynjolfsson & Aral
Statistical Models
Information practices that matter…
Page 23
© 2006 Van Alstyne, Brynjolfsson & Aral
IT variables Intermediate Output Final Output
IndividualCompensation
A Model of Information Work: Task Completion & Compensation
Revenue CompensationCompletion
Rate
Multitasking
Duration perTask
DatabaseSkill
EmailContacts
IT variables Intermediate Output Final Output
IndividualCompensation
Page 24
© 2006 Van Alstyne, Brynjolfsson & Aral
Model Specification
Qi – Output ($, Completions, Duration …)
Hi – Job Level (Partner, Consultant, Rsch …)
Xi – Human Capital (Ed., Exp., Labor)
Yi – IT Factor (Email, Ties, Behaviors…)
' 'i i i i iQ Y e H X
Page 25
© 2006 Van Alstyne, Brynjolfsson & Aral
Source | SS df MS Number of obs = 41-------------+------------------------------ F( 6, 34) = 1.33 Model | 1.9341e+11 6 3.2236e+10 Prob > F = 0.2691 Residual | 8.2136e+11 34 2.4158e+10 R-squared = 0.1906-------------+------------------------------ Adj R-squared = 0.0478 Total | 1.0148e+12 40 2.5369e+10 Root MSE = 1.6e+05
------------------------------------------------------------------------------ rev02 | Coef. Std. Err. t P>|t| [95% Conf. Interval]-------------+---------------------------------------------------------------- partner | 239727.5 141685.8 1.69 0.100 -48212.66 527667.6 consultant | 272197.7 112464.6 2.42 0.021 43642.14 500753.2 gender | -65767.58 55093.9 -1.19 0.241 -177731.8 46196.69 age | 5852.73 4143.612 1.41 0.167 -2568.103 14273.56 yrs_educ | -1842.269 23137.51 -0.08 0.937 -48863.34 45178.81 experience | 681.794 3977.229 0.17 0.865 -7400.908 8764.496 _cons | -69840.65 530698 -0.13 0.896 -1148349 1008667------------------------------------------------------------------------------
HR Factors
Source | SS df MS Number of obs = 33-------------+------------------------------ F( 6, 26) = 12.63 Model | 4.6776e+11 6 7.7959e+10 Prob > F = 0.0000 Residual | 1.6051e+11 26 6.1735e+09 R-squared = 0.7445-------------+------------------------------ Adj R-squared = 0.6856 Total | 6.2827e+11 32 1.9633e+10 Root MSE = 78572
------------------------------------------------------------------------------ rev02 | Coef. Std. Err. t P>|t| [95% Conf. Interval]-------------+---------------------------------------------------------------- icontacts | 6553.851 1804.091 3.63 0.001 2845.488 10262.21 searchtools | 204.9083 159.1239 1.29 0.209 -122.1756 531.9923 betweenness | 107.8983 43.14879 2.50 0.019 19.20467 196.5919 partner | 175545 64618.17 2.72 0.012 42720.41 308369.5 consultant | 298923.3 65735.69 4.55 0.000 163801.7 434045 multtsks | 25275.27 7197.28 3.51 0.002 10481.05 40069.49 _cons | -467132.8 165420.2 -2.82 0.009 -807158.8 -127106.7------------------------------------------------------------------------------
IT Factors
Page 26
© 2006 Van Alstyne, Brynjolfsson & Aral
A Model of Information Work: Tasks & Completion Rate
Intermediate Output Final Output
IndividualCompensation
Revenue CompensationCompletion
Rate
Multitasking
Duration perTask
Intermediate Output Final Output
IndividualCompensation
Do multitasking and duration affect completed projects ?
Page 27
© 2006 Van Alstyne, Brynjolfsson & Aral
Table 3: Panel Data Estimates of Project Completion, Revenues Multitasking, and Duration (n=NormVar) Variables Model 1 Model 2 Model 3 Model 4 Model 5 Dependent Variable:
Revenues nRevenues nRevenues nComp. Projects
nComp. Projects
Specification FGLS FGLS Fixed Effects FGLS Fixed Effects Daily Daily Daily Daily Daily
Controls
Education, Gender, Partner,
Consultant
Education, Gender, Partner,
Consultant
Education, Gender, Partner,
Consultant
Ind. Variables Completed Projects
2149.19*** (43.41)
nMultitasking
1.067*** (.005)
.707*** (.004)
.692*** (.003)
.603*** (.003)
nMultitasking2
-.108***
(.002) -.109***
(.002) -.066***
(.002) -.112***
(.001) nDuration
-.094***
(.002) -.143***
(.003) -.173***
(.003) -.149***
(.002)
Time Controls Month, Year
Month, Year
Month, Year
Month, Year
Month, Year
Log Likelihood -370966.8 194415.3 - 159337.5 - X2(d.f) / F(d.f) 8976.9***
(20) 73509.3***
(22) 5566.41***
(18) 59109.4***
(22) 3848.13***
(18) Observations 78201 78201 100816 81824 100816 ***p<.001; **p<.05; *p<.10
A worker generates $2149.19 per project, per day for the firm.
Multitasking associated with increases in completed projects & revenues.
Longer duration associated with decreases in both completed projects & revenues.
MT2 is negative, implying an inverted-U shaped relationship
What Drives Revenue Generation?
Y
MT
On average,
$CP
MT
D
Page 28
© 2006 Van Alstyne, Brynjolfsson & Aral
IT variables Intermediate Output Final Output
A Model of Information Work:
RevenueCompletion
Rate
Multitasking
Duration perTask
DatabaseSkill
EmailContacts
IT variables Intermediate Output Final Output
Do IT skills & social networks affect multitasking and duration?
Page 29
© 2006 Van Alstyne, Brynjolfsson & Aral
Multitasking and Duration depend on DB-Skill and Contact Networks
•Contact networks and DB-Skill help workers multitask •But average duration suffers.
IT Intermed
Coefficientsa
-1.769 6.223 -.284 .779
2.396 1.762 1.360 .186
2.636 2.056 1.282 .212
.126*** .043 2.941 .007
.009** .004 2.375 .026
(Constant)
Consult Dummy
Partner Dummy
Total Internal Contactsin Incoming Emails
DB_SKILL
B Std. Error
Unstandardized Coefficients
t Sig.
Dependent Variable: MULTTSKSa.
Coefficientsa
-26.821 147.052 -.182 .857
16.382 36.720 .446 .660
20.128 45.193 .445 .660
1.906* .987 1.931 .066
.169* .083 2.027 .054
B Std. Error
Unstandardized Coefficients
t Sig.
Dependent Variable: AVEDURa.
Multitasking Duration
Adjusted R2 = .24 with controls for GENDER, YRS_ED, YRS_EXP.b.
Adjusted R2 = .18 with controls for GEN., ED., and EXP.b.
Page 30
© 2006 Van Alstyne, Brynjolfsson & Aral
Multitasking, Duration and Completion Rate
Time
B
A
CompletedProjects
3
5
Page 31
© 2006 Van Alstyne, Brynjolfsson & Aral
Relation Between IT & Multitasking
F2F – small magnitude positive with MT. Interviews indicate that a certain number of
F2F meetings are necessary for each additional project.
Heavy Multitaskers rely more on asynchronous email and less on synchronous phone communication.
ESS Use positively correlated with multitasking.
Project Coordination – labor, anti-poaching Cross Project Info Seeking Need more information relevant to more
searches.
Interaction Term: Information Seeking and Information Communication are Complements in regards to MT Behavior
Page 32
© 2006 Van Alstyne, Brynjolfsson & Aral
MultitaskingAsynchronousInformation
Seeking Helps!
SynchronousInformation
Seeking Hurts!
• Email• DB Access
• Phone
© That Girl
Initial Synchronize: • Face to Face
Page 33
© 2006 Van Alstyne, Brynjolfsson & Aral
IT variables Intermediate Output IndividualCompensation
Revenue CompensationCompletion
Rate
Multitasking
Duration perTask
DatabaseSkill
EmailContacts
IT variables Intermediate Output IndividualCompensation
A Model of Information Work: Executive Recruiting Case
Final OutputFinal
Output
Page 34
© 2006 Van Alstyne, Brynjolfsson & Aral
Check: Revenue & Compensation do depend on IT Skills
The more observable contact network helps revenue and compensation.
The less observable DB-skill helps revenue but hurts compensation.
IT
Coefficientsa
(Constant)
Consult Dummy
Partner DummyTotal Internal Contactsin Incoming Emails
DB_SKILL
B Std. Error
Unstandardized Coefficients
t Sig.
Dependent Variable: REV02a.
Coefficientsa
B Std. Error
Unstandardized Coefficients
t Sig.
Dependent Variable: SALARYa.
Revenue Compensation
-333896.63 306222.69 -1.090 .286
420625.63*** 86713.60 4.851 .000
354668.03*** 101188.43 3.505 .002
11657.50*** 2102.10 5.546 .000
326.32* 194.74 1.676 .106
133654.46 152918.8 .874 .388
148254.60*** 29454.27 5.033 .000
317464.32*** 44561.70 7.124 .000
1953.29** 841.10 2.322 .026
-204.22* 116.98 -1.746 .089
Adjusted R2 = .53 with controls for GENDER, YRS_ED, YRS_EXP.b.
Adjusted R2 = .77 with controls for GEN., ED., and EXP.b.
$ Comp
Page 35
© 2006 Van Alstyne, Brynjolfsson & Aral
Recall Network Position…
Betweenness Constrained vs. Unconstrained
Page 36
© 2006 Van Alstyne, Brynjolfsson & Aral
Network Structure Matters
Coefficientsa
(Base Model)
Size Struct. Holes
Betweenness
B Std. Error
Unstandardized Coefficients
Adj. R2 Sig. F
Dependent Variable: Bookings02a.
Coefficientsa
B Std. Error
Unstandardized Coefficients
Adj. R2 Sig. F
Dependent Variable: Billings02a.
New Contract Revenue Contract Execution Revenue
0.40
13770*** 4647 0.52 .006
1297* 773 0.47 .040
0.19
7890* 4656 0.24 .100
1696** 697 0.30 .021
Base Model: YRS_EXP, PARTDUM, %_CEO_SRCH, SECTOR(dummies), %_SOLO.b.
N=39. *** p<.01, ** p<.05, * p<.1b.
Bridging diverse communities is significant.
Being in the thick of information flows is significant.
Page 37
© 2006 Van Alstyne, Brynjolfsson & Aral
Information Flows Matter
Coefficientsa
(Base Model)
Best structural pred.
Ave. E-Mail Size
Colleagues’ Ave.Response Time
B Std. Error
Unstandardized Coefficients
Adj. R2 Sig. F
Dependent Variable: Bookings02a.
Coefficientsa
B Std. Error
Unstandardized Coefficients
Adj. R2 Sig. F
Dependent Variable: Billings02a.
New Contract Revenue Contract Execution Revenue
0.40
12604.0*** 4454.0 0.52 .006
-10.7** 4.9 0.56 .042
-198947.0 168968.0 0.56 .248
0.19
1544.0** 639.0 0.30 .021
-9.3* 4.7 0.34 .095
-368924.0** 157789.0 0.42 .026
Base Model: YRS_EXP, PARTDUM, %_CEO_SRCH, SECTOR(dummies), %_SOLO.b.
N=39. *** p<.01, ** p<.05, * p<.1b.
Sending shorter e-mail helps get contracts and finish them.
Faster response from colleagues helps finish them.
Page 38
© 2006 Van Alstyne, Brynjolfsson & Aral
Do larger personal rolodexes make you more productive?
Page 39
© 2006 Van Alstyne, Brynjolfsson & Aral
H5: Recruiters with larger personal rolodexes generate no more or less output
Revenue $ $ for completed searches
Completed searches
Multitasking Duration Duration controlling
for multitasking
Size of rolodex (Q50)
-10.2 (60.3)
-22.9 (32.6)
0.000 (0.001)
0.000 (0.001)
-0.013 (0.021)
-0.013 (0.016)
• Less information sharing• Less DB proficiency• Lower % of e-mail read• Less learning from others• Less perceived credit for ideas given to colleagues• More dissembling on the phone
Instead, a larger private rolodex is associated with:
* p < 0.10, ** p < 0.05, *** p < 0.01, Standard err in paren.
Page 40
© 2006 Van Alstyne, Brynjolfsson & Aral
Interesting & Notable Correlations
Page 41
Within Survey Correlations
Across all 3 job types• Volunteering info Giving credit• Sharing Happiness• Indiv performance Objective metrics
- Supervisor input• Gathering internal/external info Happiness• Yrs Experience - public access web pages• Age Experience, Rolodex• Accurate DB Happier• Overlapping social network Effective use of phone
Significant at 10% level
Page 42
Correlations w/ Completed Job Searches
For consultants perceived accuracy of corporate DB professed ability to use internal IT support tools having control over the data accessed & used more people contacted per day• - relative time spent processing info on computer screen• - personal knowledge never entered in DB
For partners with info pull (request data not wait for it)• - procedural communication instead of descriptive info• - reporting severe costs to not having info when need it
Significant at 10% level
Page 43
Correlations w/ Multitasking
For consultants perceived accuracy of corporate DB finds more relative value in internal DB having routine data requirements happy in current job• - relative time spent on public access web pages
For partners if private info not entered in DB, main reason is too tedious
Significant at 10% level
Page 44
Correlations w/ Revenue
For Consultants number of people contacted via e-mail percent time spent on e-mail (1 firm < 0!) more relative time spent with external DB more value from internal DB• - reporting problem of info overload
For Partners individual (not team) based compensation most relative time spent with external people• - personal knowledge never entered in DB• - there are multiple sources for key info
Significant at 10% level
Page 45
© 2006 Van Alstyne, Brynjolfsson & Aral
Having IT is not enough.It’s how you use and manage information
and contacts that matters.
Page 46
© 2006 Van Alstyne, Brynjolfsson & Aral
Takeaways 1
1. We have strong evidence associating different IT practices and social networks with measures of white collar output. Social Network links are worth > $6,000 in this context.
2. Economics: incentive design mechanisms do correspond with information sharing.
3. Social network strategies are (i) bridging info pools (ii) being an info hub and (iii) career ladder => Structure matters!
4. Realize efficiencies by (i) connecting to hubs (ii) short msgs (iii) declarative information (iv) encouraging timely response from colleagues (and being prompt yourself!) => Flow matters!
5. Give information back. Data monitoring is not a sin if the principal use is to support those who provide it.
Page 47
© 2006 Van Alstyne, Brynjolfsson & Aral
Takeaways 2
6. Perceived information overload corresponds very little to actual communication flows but rather to Lower comfort with IT Longer response times from colleagues With whom you communicate
7. Certain white collar knowledge mgmt practices can be routinized. Remove or automate tedium of data capture. Most successful folks will share.
8. Consider hires for willingness to share and use IT, not just individual performance. Corollary: you may need to reward this.
9. Use IT and ESS both to support multitasking and increase speed. This helps people accomplish more work.
Page 48
© 2006 Van Alstyne, Brynjolfsson & Aral
Questions?
[email protected]
[email protected]