Distinguishing Bandwidth and Latency in Households ...
Transcript of Distinguishing Bandwidth and Latency in Households ...
Distinguishing Bandwidth and Latency in Households’ Willingness-to-Pay for
Broadband Internet Speed
Yu-Hsin Liu, Jeff Prince, and Scott WallstenOctober 19, 2018
How Do Consumers Value Key Features of Home Internet Connection?
(with a Focus on Speed)
1. WTP for bandwidth highly concave; biggest increases in value < 50 MbpsKey Takeaways
Key Takeaways
2. WTP to reduce latency from satellite to fixed levels is about $8.66.
3. Failing to account for latency causes consumers to exaggerate WTP for bandwidth.
ISPs Invest Billions & Compete on Speed
Policymakers Worry Broadband Too Slow
Bandwidth Determines Whether Policy Considers Service “Broadband”
1999
4 Mbps down1 Mbps up
2010
200 Kbps up or down
2015
25 Mbps down3 Mbps up
2017
NOI
Definition Affects Policy• Mergers
• Universal Service Allocations
• Universal Service Auctions (esp CAF II)
All have real economic consequences.
Definition Does Not Take Into Account Consumer Preferences
• Key question for all relevant policy decisions should be: “Are costs of higher speed worth the benefits?”
What Is Speed?
Colloquially, “Speed” = Bandwidth
But really…
Speed = f(bandwidth, latency)
Bandwidth:The maximum throughput of the Internet path (Mbps)
Latency:The time it takes for a data packet to make the round trip between the user’s computer and another computer (ms)
The Highway Analogy
Implies Two Additional Problems1. Consumers don’t know what “speed” is.
2. Policymakers may conflate bandwidth and latency when setting rules.
Bandwidth
Bandwidth:The value of going from dial-up to low-end broadband was likely large
However, as top-level bandwidth continues to rapidly grow, do consumers still get substantial increased value?
Latency
Latency:Variation largely due to differences in technology (fixed-line vs. satellite)
How much do consumers value the difference?
We Measure Trade-Offs of Broadband Features Using Discrete-Choice Surveys
Allows us to determine:
1. Value of incremental increases in bandwidth
2. Value of latency – in absolute terms and relative to bandwidth
3. Consequences of ignoring latency when measuring value of speed
Related Literature on Internet Demand
Studies Using Market Data:Goolsbee & Klenow (2006)Nevo et al. (2016)Ahlfeldt et al. (2016)Boik (2017)
Related Literature on Internet Demand
Studies Using Surveys and Experiments:Varian (2002)Rappaport et al. (2002)Dutz et al. (2009)Rosston et al. (2010)
We Use a Survey. Why?Allows for sufficient and exogenous variation in bandwidth and latency
Only way to include latency, which does not appear in the common market data
Why a Choice-Based Survey?Mimics the real market
Mitigates reporting inaccuracy of stated-preference data
Has become the dominant approach
Survey ChallengesMitigating hypothetical bias
Describing broadband features
Constructing choices
Four Parts of the Survey1. Demographic questions
Four Parts of the Survey1. Demographic questions
2. Home Internet activities and device use questions
Four Parts of the Survey1. Demographic questions
2. Home Internet activities and device use questions
3. Cognitive build-up to make sure respondents understand the questions
Cognitive BuildupAverageBandwidthLevels
CanStreamVideo/VideooverIPonasingledevice(StandardDefinition,SD)
CanStreamVideo/VideooverIPonasingledevice(High-Definition,HD1080P)
QualitypossibleforStreamingVideo/VideooverIPon2-4devicesconcurrently(UltraHD2160P)
Timetodownload/upload5GBfile(e.g.,2hourHDmovie)
4Mbps Yes 167mins
10Mbps Yes Yes SD 67mins
25Mbps Yes Yes HD 27mins
50Mbps Yes Yes HD 13mins
75Mbps Yes Yes HD 9mins
100Mbps Yes Yes UltraHD 7mins
150Mbps Yes Yes UltraHD 4mins
300Mbps Yes Yes UltraHD 2mins
500Mbps Yes Yes UltraHD 1mins
1000Mbps Yes Yes UltraHD 0.7mins
Cognitive Buildup
AverageLatencyLevels
OnlineGaming VoiceoverIP Web-pageloadingtimeforanaveragewebsite
(Belshe2010)
Lessthan10ms
Unnoticeabledelay Unnoticeabledelay 0.8secs
10- 30ms Unnoticeabledelay Unnoticeabledelay 1.1secs
30- 60ms Minordelay Unnoticeabledelay 1.5secs
60- 150ms Somedelay Somedelay 2.7secs
150- 300ms Significantdelay Significantdelay 4.8secs
300- 600ms MoreSignificantdelay MoreSignificantdelay 8.9secs
Four Parts of the Survey1. Demographic questions
2. Home Internet activities and device use questions
3. Cognitive build-up to make sure respondents understand the questions
4. Choice experiments
Choice Experiments• Choose among hypothetical alternatives• Compare choice to actual current plan
Choice Experiment: ExampleInternetA InternetB InternetC InternetD
Price $45 $70 $80 $35
DownloadBandwidth
25Mbps 75Mbps 1000Mbps 4Mbps
UploadBandwidth
3Mbps 3Mbps 25Mbps 1Mbps
Latency 60to150ms 300to600ms 30to60ms Lessthan10ms
DataCap 600GB 300GB 1000GB Unlimited
PickOne ○ ○ ○ ○
Choice Experiment: ExampleYourChosenPlan YourCurrentPlan
Price Reported/inferredprice
DownloadBandwidth Reported
UploadBandwidth Reported/inferred
Latency Reported/predicted
DataCap Reported/unlimted
PickOne ○ ○
Choices Imply PreferencesPreferences embed trade-offs (between price and broadband components)
Trade-offs tell us about WTP
Two SurveysOne asks about latencyOne does not
à Makes it possible to see how people conflate bandwidth and latency when discussing “speed”
Focus Group Pre-TestNo survey fatigue
Highway analogy & tables helpful, intuitive
Many don’t know all features of current broadband plan
For survey frequently had to infer relevant information
How We Infer Characteristics of Respondent’s Current Plan
Latency – based on location, ISP, plan
Data cap – assumed unlimited
Upload bandwidth – based on technology
The SurveyConducted by ResearchNow
To be included in our sample, respondent must:• Currently subscribe to home Internet• Be household primary decision-maker• Know monthly fee & download bandwidth• Take > 4 mins to complete survey
Sample sizes:• 978 households using latency survey• 433 households using non-latency survey
What People Do With Their Internet
Note: Figure shows weighted average of latency and no-latency survey.
Characteristics of Respondents’ Subscriptions
The ModelWe use standard McFadden-style utility model to estimate preferences:
Utility:
x includes Internet features
Status quo option includes one additional component…
𝑢"#$ = 𝒙′"#$𝜷 + 𝜀"#$
The ModelStatus quo utility:
Allow for distribution of preferences for status quo, per se, that varies by person
Can incorporate using random coefficient
𝑢"#$ = 𝒙′"#$𝜷 + 𝛾𝒊𝑆𝑄"#$ + 𝜀"#$
EstimationWith standard assumption for distribution of unobservables (Type I extremum), we have choice probabilities:
Part 1:
Part 2:
Prob 𝑌"$ = j =ex p( 𝒙′"#$𝜷9
∑ ex p( 𝒙′"#$𝜷9;#<=
Prob 𝑌"$ = j = >ex p( 𝒙′"#$𝜷 + 𝛾𝒊𝑆𝑄"#$9
∑ ex p( 𝒙′"#$𝜷+𝛾𝒊𝑆𝑄"#$9?#<=@
EstimationEstimate parameters using maximum likelihood
WTP estimates are simply ratios marginal utility (β’s)
WTP for download bandwidth is ratio of coefficients on download bandwidth and price:
𝑊𝑇𝑃(𝑏𝑎𝑛𝑑𝑤𝑖𝑑𝑡ℎ) = −𝛽OP𝛽Q
Results: WTP for Bandwidth
Download WTP10 Mbps $14.01***25 Mbps $37.63***50 Mbps $51.80***75 Mbps $59.70***100 Mbps $63.82***150 Mbps $71.37***300 Mbps $75.60***500 Mbps $75.47***1000 Mbps $82.59***
WTP Estimates From the 2 Surveys
Results: WTP for (Lower) LatencyLatency WTP10-30 ms $3.78**30-60 ms -$0.3660-150 ms -$4.03**150-300 ms -$6.45***300-600+ ms -$8.66***
Reducing latency from 300-600+ ms to below 60 ms is roughly equivalent to switching from satellite to fixed-line service.
WTP for Higher Data Caps
Data cap600 GB $12.21***1000 GB $23.57***Unlimited $58.20***
ResultsDownload WTP10Mbps $14.01***25Mbps $37.63***50Mbps $51.80***75Mbps $59.70***100Mbps $63.82***150Mbps $71.37***300Mbps $75.60***500Mbps $75.47***1000Mbps $82.59***Upload3Mbps $10.01***25Mbps $18.57***100Mbps $24.46***Latency10-30ms $3.78**30-60ms -$0.3660-150ms -$4.03**150-300ms -$6.45***300-600+ms -$8.66***Datacap600GB $12.21***1000GB $23.57***Unlimited $58.20***
Results Differ Across GroupsWe allow for differences in WTP across demographics and usage
Education & Income are as we’d expectNotable difference in download
bandwidth
Gamers most notably sensitive to latency
Example: CAF AuctionReverse auction to allocate $2 billion
Bids are scored as function of bandwidth and latency:
𝑠𝑐𝑜𝑟𝑒 = 100𝑠𝑢𝑏𝑠𝑖𝑑𝑦𝑟𝑒𝑞𝑢𝑒𝑠𝑡𝑒𝑑𝑟𝑒𝑠𝑒𝑟𝑣𝑒𝑝𝑟𝑖𝑐𝑒
+ 𝑝𝑒𝑟𝑓𝑜𝑟𝑚𝑎𝑛𝑐𝑒𝑤𝑒𝑖𝑔ℎ𝑡 + 𝑙𝑎𝑡𝑒𝑛𝑐𝑦𝑤𝑒𝑖𝑔ℎ𝑡
CAF Auction Bandwidth WeightsPerformanceTier Bandwidth UsageAllowance Weight
Minimum ≥10/1Mbps ≥150GB 65
Baseline ≥25/3Mbps ≥150GBorU.S.median,whicheverishigher
45
AboveBaseline ≥100/20Mbps 2TB 15
Gigabit ≥1Gbps/500Mbps 2TB 0
CAF Auction Latency Weights
Latency Requirement Weight
Low ≤100ms 0
High ≤750ms &“meanopinionscore”of≥4
25
Did the FCC Get the Weights Right?
Performance and Latency Weights Should Reflect Consumer Valuation
Bandwidth Weights Reflect Decreasing WTP
Bandwidth Scoring Not Bad!Performance
TierScore Marginal
Reduction in Score
Change relative to previous change
Marginal Increase in WTP
WTP change relative to previous change
Implied WTP per
Point
Minimum 65 . .
Baseline 45 20 $33.37 $1.67Above Baseline 15 30 1.5 $56.73 1.7 $1.89
Gigabit 0 15 0.5 $36.32 0.6 $2.42
Too Much Weight on LatencyPerformance Tier
Reduction in Score From Latency Improvement
Increase in WTP From Latency Improvement
FCC Latency Weight Relative to
WTPMinimumBaseline 1.25 0.26 4.8Above Baseline 0.83 0.15 5.5Gigabit 1.67 0.24 7.0
Latency Tier Points Point Reduction
Increase in WTP
WTP/Point
High 25Low 0 25 $8.66 $0.35WTP/Point for Bandwidth $1.67 - $2.42
TakeawaysValuation of bandwidth is highly concave
Households moderately value latencyExcluding it exaggerates bandwidth value
Without this information policies may inadvertently under- or over-weight characteristics that consumer care about.
FIN
Outline of TalkSpeed & Its Relation to Policy
Survey Design
Data
Estimation & Findings
Takeaways
Speed & PolicySpeed:
Bandwidth:DownloadUploadEvolving FCC benchmarks
LatencyNo benchmark but FCC measurement
began in 2011 (Measuring Broadband America)
Speed & PolicyWhat is Speed?
From a user’s perspective = time it takes to complete a meaningful task
Viewed this way, both bandwidth and latency play a role
Each can be key component to the taskThink video streaming vs. gaming
Speed & PolicyPolicy has typically focused on bandwidth
It is also the focus in private competition
Several concerns arise1. Are policies focused on bandwidth maximizing consumer welfare?
2. How do we incorporate latency into policy?
3. We know little/nothing about how consumers value bandwidth/latency
Representative SampleDemographics
Region, age, race, sex, education, employment status, household income
Representative sample on:RegionAgeRaceSex
Survey DesignCognitive Build-up
Feature descriptions
Follow by asking about:Connected devices at homeUses of home InternetCurrent home Internet plan
Survey DesignChoice experiments
Respondents make 8 choicesTwo parts:
Choose among hypothetical alternatives
Compare choice to current plan
Omitted features assumed identical across alternatives