Date post: | 10-Nov-2023 |
Category: |
Documents |
Upload: | independent |
View: | 0 times |
Download: | 0 times |
Information Overload and Virtual Public Discourse Boundaries
Quentin Jones, Gilad Ravid & Sheizaf Rafaeli Graduate School of Business University of Haifa,
Mount Carmel, Haifa 31905, Israel [email protected]
Abstract: ‘Virtual publics’ are a type of computer mediated discourse space created by using various technologies including email, the USENET, web based bulletin boards, IRC, MUDS, etc. (Jones and Rafaeli 1999). This paper outlines on-going field research into the impact of information overload on virtual public “mass interaction”. It describes initial findings based on an examination of 2.65 million USENET messages, which suggest that a number of user ‘information overload’ coping strategies appear to impact on the overall shape of group-discourse. The research in progress outlined addresses issues associated with group level usability of computer mediated communication technologies. Keywords: Virtual Communities, Virtual Publics, Computer Mediated Communication, Information Overload. 1 Introduction The importance of virtual communities is widely accepted (Jones and Rafaeli 2000b). However, no single dominant definition of the term exists (Jones 1997). In fact, a number of authors dispute their existence (Jones and Rafaeli 2000b). Consequently, despite the significance of the phenomena commonly labeled virtual community, the term is problematic. This is because the term as currently understood does not clearly distinguish between online places; computer mediated interactions; and community arising from or related to such interactions (Jones 1997; Jones and Rafaeli 2000b). Once these terms are distinguished, then in part, the question of whether a virtual space is associated with a virtual community becomes an empirical question (Liu 1999). To avoid this potential confusion the term virtual public has been proposed. Virtual publics are symbolically delineated computer mediated spaces, whose existence is relatively transparent and open, that allow groups of
individuals to attend and contribute to a similar set of computer-mediated interpersonal interactions.
The value of any virtual public will relate to the size of its user population, and both the quality and quantity of user contributions. Unfortunately, virtual public expansion is not simply a matter of increasing user population. This is because the relationship between user population and discourse contributions is influenced by a variety of factors including, critical mass, social loafing and information overload (Jones and Rafaeli 1999). While these points can be deduced from existing empirical and theoretical works, little data has been presented that concretely describes how these phenomena are related to various technologies. This research addresses this issue.
2 Modeling the Connection Between Communication Technology and Discourse
The degree to which information technologies can effectively control or aid computer mediated communication (CMC) is limited by the finite
capacity of human cognition. It follows that the inability of users to process effectively certain message patterns will result in limitations to the
possible forms of sustainable group-CMC. That is, beyond a particular communication-processing load, the behavioral stress zones encountered will make particular forms of group communication unsustainable. Communication load is the processing effort required to deal with a set of communications.
A
B
C
Members of a Virtual Public
Cog
niti v
e P r
oces
s ing
Load
for G
rou p
-CM
C
Variable Levels ofCognitive Effort
Individuals are Willing toInvest in Processing
Interactive CMC
Figure 1: Cognitive Load and Group-CMC
Figure 1 illustrates how individual cognitive processing limits are linked to group communication. Three cognitive processing loads for group-CMC, A, B and C are displayed. At different cognitive loads, different individuals are willing or able to process group messages. The cognitive processing abilities of groups are not simply the sum of its individual’s cognitive processing capacities. Consequently, certain patterns of interactive group-CMC cannot be sustained if the required processing effort (communication load) is higher than the maximum amount individuals can or are prepared to invest. In cases of “mass interaction” (Whittaker, Terveen, et al. 1998), such as Usenet discourse, in theory it should be possible to observe empirically the combined effects of the average maximum communication load individuals are prepared to process (AvMaxCL). In other words, mass interaction provides a unique opportunity to explore the impact of communication load on group discourse. This is because the large numbers involved allow broad statistical processes to be observed that may otherwise be hidden by differences between individuals and the social contexts of communication.
We know that communication-processing load relates to a number of message-system characteristics. Users generally have to make more of an effort to reply coherently to a thread (Lewis and Knowles 1997) than to a single message. Therefore, higher interactivity correlates with higher communication-processing load. Interactive communication refers here to the extent to which
messages in a sequence relate to each other, and especially the extent to which later messages recount the relatedness of earlier messages (Rafaeli and Sudweeks 1997). Similarly, a high frequency of postings will require more processing by group members. Therefore, message frequency will also covary with communication-processing load. It is also likely that a decrease in ‘interactional coherence’, not compensated for by a useable persistent record, will increase communication-processing load (Herring 1999). For example, disrupted turn adjacency may require increased user effort to track sequential exchanges. Disrupted turn adjacency is caused by the fact that CMC-systems, such as email lists, transmit messages in the order they are received. Thus in group-CMC a message may be separated from the previous message it is responding to by another message, or lags in message transmission may even result in reversed sequencing.
Different CMC-technologies have different typical message system characteristics. Therefore, the point at which a user population’s interactions will typically result in information overload will relate to the CMC-tool used.
The above discussion suggests that the potential number of users involved in ‘a sustained pattern of interactive group discourse’ via a single virtual public will be constrained by cognitive processing limits. Further, that such limits will produce communication stress zones, or boundaries, related to: message posting frequency; average depth of threads; the quantity of discourse material; disrupted turn adjacency; and how the technology in question structures communication. These communication boundaries are not rigid, deterministic, instantaneous halt lines. Rather, they are zones, with an uncertain range of likelihood within which the behavioral limitations become severe. The research described here aims to identify such cognitive processing limits induced boundaries as they relate to Usenet discourse.
3 Cognitive Processing Limits Model and Usenet Discourse
The study aims to identify the stress zones produced by cognitive processing limits for USENET newsgroups by the analysis of data collected from large-scale field research.
From the introduction above, it can be concluded that a Usenet newsgroup’s communication load will roughly be an unknown function of: Probable Major Contributors
1. Number of Posters 2. Number of Posts 3. Number of Threads 4. Message Response Times 5. Depth of Threads 6. Length of Messages 7. Discourse Type 8. Probable Moderate to Minor Contributors 9. Extent of Interactional Incoherence 10. Percent of posts / posters that are Broadcast 11. Percent of posts / posters that are Reactive 12. Percent of posts / posters that are
Interactive The model derived from Communication Load
function (CLf) leads to testable predictions about how these items interrelate. Its logic is as follows: - as the number of interactive posters should logically positively correlate with: 1) the number of interactive messages and 2) the response times to postings; the only major contributors to the CLf that can vary freely are thread depth, interactive message length, and the number of threads. Therefore, the CLf predicts that at AvMaxCL an increase in the number of interactive posters will typically be associated with one of a number of compensatory strategies or effects that operate on these variables. These strategies include: o Making an increased effort (short term solution
only, not examined in this study). o Learning new information management
techniques to reduce information overload. This is mostly relevant for inexperienced users of a particular technology. For example, Hiltz and Turoff (1985) found that feelings of overload would peek at intermediate levels of CMC use when communication volume has built up but users have not had time to develop screening skills, Herring (1999) found similar effects in relation to IRC (short term solution, not examined in this study).
o Failing to respond to certain messages, thereby lowering the growth in communication load. The result being a change in the relationship between user-population and contributions.
o Producing simpler responses, although to maintain effective communication such an approach will be asymptotic. This will result in a change in the form group communication takes
o Storing inputs and responding to them as time permits. Again changing the group communication pattern (not examined here).
o Make more erroneous responses (not examined in this study).
o Ending/reducing active participation in the group communication (Finholt and Sproull 1990).
All of these strategies have a potentially observable effect on mass interaction although changes in users’ expertise and effort are likely to produce only short-term effects. This study examines three of these strategies: 1) failing to respond to certain messages; 2) producing simpler responses; and 3) ending or reducing active participation. The AvMaxCL model makes predictions about the impact of these strategies on mass interaction, which is that their adoption would generally involve a reduction in one or more of the following variables: average thread depth; the length or complexity of messages (e.g. the number of lines or words); number of simultaneous threads; and user population turn-over. Figure 2 below is used to explain this point.
Virtual Public Technology
and Message Processing Capacity
Interactive Users-Population
Virtual Public Technologies
Hypothetical Plots of Usenet Discourse
Hypothetical Plots of Asynchronous CMC-Tool Discourse
Free
Var
iabl
es
Hypothetical AvMaxCL or I-Limit For Usenet
Figure 2: Hypothetical Model of Overload and Group-CMC
Figure 2 (based on theoretical model presented at
HICSS; Jones and Rafaeli 2000a) illustrates the expected relationship between the number of interactive users and the thread depth and length of messages. The figure also aims to highlight the importance of such a mapping process by graphically illustrating how we can come to understand differences in the group level usability of various CMC-tools. The AvMaxCL line or I-limits zones represents the boundaries to interactive communication for a particular technology.
Discovering how the boundaries differ between technologies is of great importance. A variety of interesting questions are also raised about the relationship of different kinds of discourse to such boundaries. To test the prediction described above this study mapped the range of naturally occurring patterns of sustained interactive communication associated with mass interaction via the Usenet.
It can be concluded from this model that if the structure of a newsgroup’s discourse is close to AvMaxCL, then users will try to offset an increase in any one part of the function by decreases in other parts. However, this is complicated by the fact that most of the variables listed above are highly inter-related. For example, an increase in the number of interactive posters should correlate positively with an increase in the number of interactive messages. Despite this, we should still be able to observe some general effects such as longer messages being less likely to start discussion threads in mass interaction.
The model above leads to the following testable hypotheses:
1) That longer more overloaded messages would be less likely to seed new discussions (fail to respond strategy);
2) That up until asymptote, larger more overloaded groups will have shorter messages (simpler response strategy);
3) That there will be a richness-versus-reach tradeoff (Evans and Wurster 1997) in high volume discourse (simpler response strategy). The existence of the number of threads/thread depth balance is in part counterintuitive, as one would expect the average depth of interactive discussion threads to increase along with an increase in the number of distinct interactive posters. The model in fact allows for this, but only below AvMaxCL. At AvMaxCL, one would expect a compensatory strategy to be adopted. If the compensatory strategy is not an option then other coping strategies will typically be employed.
4) As group size increases the stability of group membership will decrease (Ending/reducing active participation strategy).
4 Research Methodology
4.1 Data Collection and Sampling USENET -: The USENET is a system of electronic bulletin boards, referred to as newsgroups. It is not a computer network, but rather a network of bilateral
agreements among system administrators to cooperate on bulletin board management (Sproull and Faraj 1997).
Representative sampling of USENET discourse is difficult (Rafaeli and Sudweeks 1998). Whittaker et al’s (Whittaker, Terveen, et al. 1998) solution was to produce a randomly stratified sample. They extracted 500 newsgroups from a subset of then active, widely distributed newsgroups, which contained mostly conversational messages.
For this project, data was collected from the 500 newsgroups studied by Whittaker et al enabling detailed historical comparisons. An additional 100 newsgroups were also examined to allow for the exploration of 100 moderated and 500 unmoderated newsgroups. These additional newsgroups were selected using Whittaker et al’s approach with only minor modifications. This allowed for more moderated groups to be selected. Over a 6-month period, 2.65 million USENET messages were collected.
5 Results
5.1 Descriptive Statistics Table 1, displayed in appendix one, shows overall and comparative Usenet ‘demographic’ statistics. This table gives an inkling of the type data that has been collected and the types of analysis possible. The table also highlights the some of the likely differences in the structure of discourse between moderated and unmoderated discourse, and various discourse types. For example, 46% of one-way messages to computer focused Usenet groups resulted in a response, while only 31% of one-way messages to social discourse Usenet newsgroups resulted in a response.
5.2 Cognitive Processing Limits Model For the cognitive processing limits model to be coherently tested, multivariate time-series statistical techniques need to be applied to the data. Because of the complexity of the data set, the authors are currently attempting to conduct such modeling via various machine-learning techniques. At present, the results of such modeling are not available. However, initial overly simplistic analysis does look encouraging, and is presented below. All figures and statistics presented below are based on analysis of the entire data set of 2.65 million messages. Data plots are aggregated daily totals for each newsgroup, except for user turnover where a monthly aggregation is used.
Hypothesis 1, that longer more overloaded messages would be less likely to seed new messages, (fail to respond strategy) appears to be confirmed. Five measures of message length were examined as describe in Table 2 via T-Tests comparing seeding and non-seeding one-way messages. A message was considered one-way if a parent message for the message in question could not be found posted to its newsgroup. A seed message was one that resulted in reply. For all measures of message length examined seeding messages were found to be significantly shorter at the p<. 000 level.
SEEDS=1 N Mean 0.00 781804 49.21 Avg. No. Lines
Clients 1.00 367757 28.01 0.00 781804 44.38 Avg. No. Lines 1.00 367757 26.10 0.00 781804 34.45 Avg. No. New
Liens 1.00 367757 21.79 0.00 781804 274.98 Avg. No. Words 1.00 367757 169.55 0.00 781804 214.00 Avg. No. New
Words 1.00 367757 141.85
Table 2: One-Way message Length and Seeding
Discourse
Hypotheses 2, that up until asymptote, larger more overloaded groups will have shorter messages is examined in Figure 3 below. Figure 3 displays a scatter plot of the number of individuals that posted interactive messages (thread depth greater than 2) on a single day to a single newsgroup, against the average number lines of text contained in these messages. The measure of message length was determined by counting the number of lines of text contained in each message after dropping non-text attachments. It shape suggests that the hypothesis is correct. Plots for all measures of average message length by number of interactive posters looks similar.
Interactive Posters
3002001000-100
Avg.
No.
Inte
ract
ive
Mes
sage
Lin
es -
Con
tent
1400
1200
1000
800
600
400
200
0
-200
Figure 3. Lines in Messages by Number of Interactive
Posters
The hypothesis 3, regarding a richness-versus-reach tradeoff in thread depth is examined in Figures 4, 5, 6, and 7. All of these plots are supportive of the hypothesis.
In Figure 4, the measure of average thread depth was determined by using the dataset to recreate message threading (Or Message Chain). Using this technique each message was then assigned a depth. In this way the context of the message was taken into account.
Figure 4. Thread Depth by Chain by Number of
Interactive Posters
In Figure 5, the measure of average thread depth was determined by examining each message header and content to estimate its thread depth. Variables used to determine thread depth in this manner included: the number of message-id’s in the references section of the header; the number of Re:’s in the subject; the number and extent of message indenting of message content.
Figure 5. Thread Depth by Message by Number of
Interactive Posters
In Figure 6, the measure of average thread depth was determined by dividing the number of interactive messages posted that day by the number of interactive threads posted to on that same day. The number of interactive threads was determined by using the data set to recreate message threads.
Figure 6. Number of Responses to Threads per Day by
Number of Interactive Posters
In Figure 7, the inter-relationship between the number of simultaneous threads and thread depth is examined. It suggests that conversational strategy can be used to cope with information overload. Discourse is either broad and shallow or deep and narrow.
Threads (Broadcast, Reactive, Interactive)
5004003002001000-100
Inte
ract
ive
Thre
ad D
epth
by
Cha
in
40
30
20
10
0
Figure 7. Thread Depth by Chain by Number of Number
of Threads.
The final hypothesis, that as group size increases the stability of group membership will decrease (ending/reducing active participation strategy), is examined below in Figure 8 and by the use of correlation measures.
Membership Stability
Messages Posted to Group
6000050000400003000020000100000-10000
Prop
ortio
nal M
embe
rshi
p
120
100
80
60
40
20
0
-20
Figure 8. User Turn Over.
Figure 8 displays the number of messages posted to the 578 newsgroups that were active during the first 5 months of the study. Proportional membership is the percent of posters per-newsgroup per-month, who also posted to the next study month. On average only 11.5% of posters sent messages 2 months in a row. The drop in the proportion of individuals involved in sustained discourse is quite strong with a Spearman's rank correlation coefficient of -.42 (p < .000).
6 Discussion At this stage while it is too early to draw any strong conclusions about the impact of communication load on Usenet discourse, initial analysis is encouraging. This is primarily because initial findings do suggest support for the four hypotheses generated from the model. This is important because it suggests that via a direct examination of mass interaction, rather than by a focus on users and their social contexts, an understanding can be gained of the impact of information overload on discourse dynamics.
This work in progress is of significance for two reasons. First, it should pave the way for comparative studies of the boundaries to sustainable interactive group discourse using a variety of CMC-tools. The authors have collected email list data for this purpose. Such comparative studies should help us start to come to terms with the link between types of CMC-tools, discourse, and discourse structure. This in turn will provide us with useful usability information at the group level. Second, the study offers hope that in the future, with sufficient modeling of the data obtained from mass interaction, we will be able to provide a formula that roughly estimates communication load in a variety of situations.
References Evans, P.B. and Wurster T.S., (1997) Strategy and the
New Economics of Information, Harvard Business Review, Sept-Oct, p.74.
Finholt, T. and L. S. Sproull (1990). “Electronic groups at work.” Organization Science 1(1): 41-64.
Herring, S. C. (1999) Interactional coherence in CMC. In Proceedings of the 32nd Hawaii International Conference on System Sciences, (Hawaii), IEEE Press.
Hiltz, S. R. and Turoff, M. Structuring computer-mediated communication systems to avoid information overload. Commun. ACM 28,7, 1985
Jones, Q., and Rafaeli, S. (2000a) What do virtual 'Tells' tell? Placing cybersociety research into a hierarchy of social explanation. In Proceedings of the 33rd Hawaii International Conference on System Sciences, (Hawaii 2000), IEEE Press.
Jones, Q., and Rafaeli, S. (1999) User population and user contributions to virtual publics: A systems model. In proceeding of Group99, (Phoenix, AZ, 1999) ACM Press.
Jones, Q. (1997) Virtual-communities, virtual-settlements & cyber-archaeology: A theoretical outline. Journal of Computer Mediated Communication, 3, 3, 1997. http://www.ascusc.org/jcmc/vol3/issue3/jones.html
Jones Q., & Rafaeli. S., (2000b) "Time to Split, Virtually: ‘Discourse Architecture’ and ‘Community Building’ as means to Creating Vibrant Virtual Publics". Electronic Markets, The International Journal of Electronic Commerce and Business Media. Vol. 10, No. 4, 2000, Routledge, London.
Lewis, D., and Knowles, K. (1997) Threading electronic mail: A preliminary study. Information Processing and Management 33,2, 1997, 209-217.
Liu, G. (1999) Virtual community presence in Internet relay chatting. Journal of Computer-Mediated Communication 5, 1, http://www.ascusc.org/jcmc/vol5/issue1/liu.html
Online September (1999): http://www.lsoft.com
Rafaeli, S., and Sudweeks, F. (1997) Networked interactivity. J.C.M.C. 2, 4, http:// www.ascusc.org/jcmc/vol2/issue4/rafaeli.sudweeks.html
Rafaeli, S., Sudweeks, F., et al. (1998) Appendix: ProjectH Overview. In Network netplay. Eds Sudweeks, F., et al., American Ass. for AI, Menlo Park CA.
Sproull, L., & Faraj, S. (1997) Atheism, sex, & databases. In Culture of the Internet. Ed. Kiesler S., Lawrence Erlbaum Ass. Mahwah, New Jersey.
Whittaker, S., Terveen, L. et al. (1998) The dynamics of mass interaction. In Proceedings of CSCW 98, (Seattle), ACM Press.
A
ppen
dix
One
Sum
mar
y D
ata
Tota
l W
hitta
ker’
s Add
ition
al M
oder
ated
Unm
oder
ated
C
omp.
M
isc.
Soc.
R
ec.
Sci.
Ta
lk.
Per-
New
sgro
up
577
493
84
80
497
234
34
76
175
50
8 To
tal N
o. P
oste
rs p
er N
ewsg
roup
48
3514
4428
38
4067
6 19
701
4638
13
1885
57
2950
2 47
163
1832
90
2417
6 10
826
No.
of O
ne-O
ff Po
ster
s per
New
sgro
up
2757
5925
2531
23
228
11
464
2642
95 1
1619
8
1656
5
2591
6
9724
1
1461
5
5224
N
o. o
f Mes
sage
s fro
m R
epea
t Pos
ters
23
6745
121
6474
5 20
2706
97
172
2270
279
5498
70
2012
11
3436
68 1
0641
11
8306
3 12
5528
Pe
rcen
t Mes
sage
s fro
m R
epea
t Pos
ters
89
.39
89.3
8 89
.55
89.4
389
.39
82.3
9 92
.18
92.7
9 91
.45
84.9
6 95
.87
No.
Bro
adca
st /
One
way
mes
sage
s 92
0879
8437
50
7712
9 45
729
8751
50
2537
05
7086
8 12
1149
40
9397
35
817
2994
3 N
o. S
eed
Mes
sage
s – st
art t
hrea
d 35
7150
3283
27
2882
3 99
4034
7210
11
7872
25
432
3764
3 15
2452
13
807
9944
N
o. M
essa
ges w
ith a
ttach
men
ts
2458
922
439
2150
27
824
311
6900
18
76
2368
12
119
788
538
No.
Rep
ly M
essa
ges –
via
con
tent
20
5654
818
7552
8 18
1020
78
926
1977
622
4939
33
1750
01
2944
77
8921
36
7740
2 12
3599
N
o. R
eact
ive
Mes
sage
s 70
4740
6449
34
5980
6 59
806
6812
90
2008
62
5815
1 75
299
3138
97
2706
4 29
467
No.
Inte
ract
ive
Mes
sage
s 10
2270
793
3270
89
437
3947
598
3232
21
2819
89
262
1739
09
4403
18
3489
7 71
502
Avg
. No.
Mes
sage
s per
new
sgro
up
4589
.82
4912
.69
2694
.9
1358
.18
5110
285
2.08
642
0.03
487
3.12
664
9.21
195
5.56
16
364
Num
ber o
f Mes
sage
s 26
4832
624
2195
4 22
6372
10
8654
2539
672
6673
86
2182
81
3703
57 1
1636
12
9777
8 13
0912
M
essa
ge L
engt
h –
cont
ent
43.9
36.9
6 88
.48
103
34.3
8 35
.21
53.7
1 56
.68
47.7
8 44
.3
47.4
1 M
essa
ge L
engt
h –
head
ers
52.4
46.2
7 84
.59
107
43.6
3 37
.6
56.9
4 60
.07
69.9
2 46
.29
49.2
5 N
umbe
r of M
essa
ges C
ross
pos
ted
4918
0444
6704
45
100
1036
748
1437
13
0387
23
047
1549
40
1029
92
3004
3 50
395
Avg
. Thr
ead
Dep
th b
y M
essa
ge
1.55
1.62
1.
17
0.8
1.68
1.
4 1.
3 2
1.49
1.
6 3.
9 A
vg. T
hrea
d D
epth
by
Cha
in
1.2
1.26
0.
84
0.54
1.31
1.
08
0.96
1.
41
1.28
1.
08
2.69
A
vg. I
nter
activ
e Th
read
-D b
y M
sg.
3.51
9997
3.53
9623
9 3.
3545
71
3.13
5582
3.55
0656
03 3.
4823
03 3.
4110
23 4.
0176
36 3.
3016
33 3.
5262
39 5.
0035
12
Avg
. Int
erac
tive
Thre
ad-D
by
Cha
in
3.31
3968
3.32
4296
3.
2269
18
3.10
164
3.33
0902
45 3.
1229
06 3.
1309
23 3.
6286
84 3.
4353
78
3.22
5 4.
465
Perc
ent o
f mes
sage
s one
-way
34
.772
1134
.837
573
34.0
7179
42
.086
8134
.459
1743
38.0
1473
32.4
6641
32.7
1141
35.1
8329
36.6
3094
22.8
7262
Pe
rcen
t of o
ne-w
ay =
seed
s 38
.783
5938
.912
830
37.3
6986
21
.736
7539
.674
3415
46.4
6026
35.8
8644
31.0
7166
37.2
3818
38.5
4873
33.2
0976
Pe
rcen
t of m
essa
ges r
eplie
s 77
.654
6477
.438
630
79.9
657
72.6
3976
77.8
6918
94 74
.010
09 80
.172
35 7
9.51
17 76
.669
54 79
.160
96 94
.413
80
Perc
ent o
f mes
sage
s rea
ctiv
e 26
.610
7726
.628
664
26.4
1934
55
.042
6126
.825
9051
30.0
9683
26.6
4043
20.3
3146
26.9
7609
27.6
7903
22.5
0901
Pe
rcen
t of m
essa
ges I
nter
activ
e 38
.617
1138
.533
762
39.5
0886
36
.330
9238
.714
9207
31.8
8844
40.8
9316
46.9
5712
37.
8406
35.6
9003
54.
6184
Pe
rcen
t of M
essa
ges w
ith A
ttach
men
ts 0
.928
473
0.92
6483
3 0.
9497
64
0.25
586
0.95
7249
60 1.
0338
84 0.
8594
43 0.
6393
83 1.
0414
98 0.
8059
07 0.
4109
63
Perc
ent o
f Mes
sage
s Cro
ss p
oste
d 18
.570
418
.443
95
19.9
230
9.54
130
18.9
5666
0 19
.537
0 10
.558
4 41
.835
3 8.
8511
30.
7257
38.
4953
Tabl
e 1:
Sum
mar
y of
Dat
a by
New
sgro
up –
St
udy
Col
lect
ion
Dat
es S
epte
mbe
r 1, 1
999
- Fe
brua
ry 2
9th 2
000