TVB Research

1
TVB Research
Conference
October 26, 2000
2
THAT PASSIVE METER DOWN THE BLOCK.
one
3
BACKGROUND
4
For more than two decades we have been trying to
build the passive TV meter.
5
The irony is there are 15,000 passive meters
operating in the US right now.
6
Theyre called set meters.
7
And though they are eclipsed by peoplemeters . .
.
8
They are and will remain the dominant local
metering technique.
9
The problem with set meter is the diary.
10
To solve this, Nielsen is moving choc-a-block to
peoplemeters in a Boston test.
11
A clear alternative is to increase the set meter
panel and model viewers.
Set meters
12
THE SET METER REVISITED
13
A set meter can be thought of as a people-meter
that doesnt measure people.
14
By doing less, it can do more.
15
It does not require the cooperation of all
household members.
16
It is totally passive . . .
17
It has fewer response problems. . .
18
It costs far less to operate.
19
But set meters do not measure viewing so that
information needs to be obtained elsewhere.
20
I believe it is quite possible to model program
viewing . . .
21
. . . from set tuning data and independent VPVH
estimates . . .
22
. . . with results indistinguishable from
peoplemeter viewing data.
23
VIEWER MODELING
24
Modeling viewers is not a new idea.
25
It was first suggested by Ehrenberg and Twyman in
1966.
26
Who argued against repeatedly measuring behavior
that shows little variation (i.e. VPS).
27
Modeling was revisited by Kirkham in 1993,
related to the successful UK TV Span set meter
panel.
28
He reported that 70 of the variation in viewer
ratings was explained by household tuning.
29
In the US viewer modeling has been demonstrated
by Ephron and Gray.
30
Who recently received ARF funding (60,000) to
perfect the viewer model.
31
It is relatively simple to model viewers because
we know a lot about what is going on in set meter
homes.
32
We know the demos of everyone in the house- hold,
the time of viewing, the set used and the program
tuned.
33
If the set in the childs room is tuned to the
Cartoon Channel, the child is likely viewing.
34
If its Oprah on the kitchen set, its likely
woman.
35
If its NFL football in the family room, its
most likely the man.
36
But the key insight is variation in VPVH for a
viewer demo will be reflected by . . .
37
. . . variation in the demo composition of the
tuned household group.
38
A high Male 18-34 VPVH will be signaled by . . .
39
. . . a high proportion of tuned households with
a Male 18-34 in residence.
40
Since the household demo comp will vary by
program, by time period, and by station . . .
41
It ties the model to real differences in VPVH for
local programs like News, Syndication and Sports.
42
THE MODEL
43
The following diagram shows the steps in modeling
viewing from household tuning data.
44
The demo Adults 35 The program 60 Minutes
45
Estimating Adults 35 60 MINUTES 1. Tuned
Households (set meter) Yes (Go to 2)
46
Estimating Adults 35 60 MINUTES 1. Tuned
Households (set meter) Yes (Go to 2)
2. With 35 adult resident? (set meter)
Yes (Go to 3) No
(Discard)
47
Estimating Adults 35 60 MINUTES
1. Tuned Households (set meter) Yes
(Go to 2) 2. With 35 adult resident? (set
meter) Yes (Go to 3)
No (Discard) 3. One person household? (set
meter) Yes (Add to viewers) No (Go
to 4)
48
Estimating Adults 35 60 MINUTES 1. Tuned
Households (set meter) Yes (Go to 2)
2. With 35 adult resident? (set meter)
Yes (Go to 3) No
(Discard) 3. One person household? (set meter)
Yes (Add to viewers) No (Go to
4) 4. Estimate probability of 35 adult
viewing in remaining 35 adult households.
(model) (Add to viewers.)
49
Estimating Adults 35 60 MINUTES 1. Tuned
Households (set meter) Yes (Go to 2)
2. With 35 adult resident? (set meter)
Yes (Go to 3) No
(Discard) 3. One person household? (set meter)
Yes (Add to viewers) No (Go to
4) 4. Estimate probability of 35 adult
viewing in remaining 35 adult households.
(model) (Add to viewers.) 5. Sum total
viewers.
50
Thats the model. Here is a demonstration using
live data.
51
Set meter data was taken from a random third A of
the NTI peoplemeter panel.
C
A
B
52
Estimating Adults 35 60 MINUTES 1. Tuned
Households
Set meter
11,465,000
53
Estimating Adults 35 60 MINUTES 1. Tuned
Households 2. With 35 adult resident?

11,465,000
Set meter
10,671,000
54
Estimating Adults 35 60 MINUTES 1. Tuned
Households 2. With 35 adult resident?
3. One person household?
11,465,000
10,671,000
Set meter
2,499,000
55
Estimating Adults 35 60 MINUTES 1. Tuned
Households 2. With 35 adult resident?
3. One person household? 4. Estimate
probability of 35 adult viewing in remaining
35 adult households.
11,465,000
10,671,000
2,499,000
56
Estimating Adults 35 60 MINUTES 1. Tuned
Households 2. With 35 adult resident?
3. One person household? 4. Estimate
probability of 35 adult viewing in remaining
35 adult households.
11,465,000
10,671,000
2,499,000
Modeled
VPVH 1.24 10,173,000
57
The demo VPVH estimate for 2 member households
with an Adult 35 in residence . . .
58
. . . uses peoplemeter data from the B third of
the sample.
59
Estimating Adults 35 60 MINUTES 1. Tuned
Households 2. With 35 adult resident?
3. One person household? 4. Estimate
probability of 35 adult viewing in remaining
35 adult households.
11,465,000
10,671,000
2,499,000
VPVH 1.24 10,173,000
5. Sum total 35 adult viewers.
12,672,000
60
Estimating Adults 35 60 MINUTES 1. Tuned
Households 2. With 35 adult resident?
3. One person household? 4. Estimate
probability of 35 adult viewing in remaining
35 adult households.
11,465,000
10,671,000

2,499,000
VPVH 1.24 10,173,000

5. Sum total 35 adult viewers.
12,672,000
61
Estimating Adults 35 60 MINUTES 1. Tuned
Households 2. With 35 adult resident?
3. One person household? 4. Estimate
probability of Male 18-49 viewing in remaining
35 adult households.
11,465,000
10,671,000
Modeled VPVH 1.11
2,499,000
VPVH 1.24 10,173,000
5. Sum total 35 adult viewers.
12,672,000
62
To validate the model, this estimate based on the
A and B thirds of the NTI sample . . .
63
Is compared to the peoplemeter estimate produced
by the C third of the NTI sample.
64
The difference is 2.
65
VPVH
Modeled 1.11 Peoplemeter 1.09
Difference 2
66
A second comparison puts a 2 difference into
perspective.
67
Two new random half-samples produce peoplemeter
VPVH estimates of 1.06 and 1.12.
68
A difference of 6.
69
VPVH. Two Half Samples
Peoplemeter A 1.06 Peoplemeter B
1.12 Difference 6
70
Here are similar comparisons for another five
randomly selected prime time programs.
71
The Practice (A18-49)

• modeled VPVH 0.69
• actual VPVH 0.68
• new split 0.72 vs. 0.66
• 1 vs. 8

72
Mon. Night FB (M18-49)

• modeled VPVH 0.42
• actual VPVH 0.38
• new split 0.37 vs. 0.43
• 11 vs. 16

73
Ally McBeal (W18-49)

• modeled VPVH 0.48
• actual VPVH 0.48
• new split 0.43 vs. 0.47
• 0 vs. 9

74
West Wing (A25-54)

• modeled VPVH 0.56
• actual VPVH 0.64
• new split 0.59 vs. 0.67
• 14 vs. 14

75
Buffy (P12-34)

• modeled VPVH 0.58
• actual VPVH 0.64
• new split 0.61 vs. 0.56
• 10 vs. 9

76
The modeled Viewer estimates are well within the
sampling error range of a 2,500 household panel .
. .
77
And are statistically indistinguishable from
measured data.
78
The next step is to apply these modeling
techniques to local demo audiences.
79
But here the smaller local samples will produce .
. .
80
. . . more variation in VPVH than the modeling
will.
81
CONCLUSION
82
Large set meter panels together with viewer
modeling . . .
83
. . . promise to produce better ratings for the
dollars spent than the current system.
84
Larger set meter panels can augment people-meter
panels for national ratings.
85
In local set-metered markets, viewer modeling can
replace diaries . . .
86
. . . which will produce better data and solve
the sweeps problem.
87
Because of costs and response problems, a
peoplemeter panel can barely measure television.
88
A set-meter panel and viewer modeling can do it
better and for less.