610 likes | 779 Views
Overview. Structure of the Experiments SP Methodology Analysis of data Development of discrete choice models Willingness-to-pay results. Attributes Tested for Domestic Customers. Attributes Tested for Business Customers. Cost also included in each experiment.
E N D
Overview • Structure of the Experiments • SP Methodology • Analysis of data • Development of discrete choice models • Willingness-to-pay results
Cost also included in each experiment • Cost levels were specified as % changes to the DUOS component of the customer’s annual electricity bill • Experiment 1: 9 cost levels, from +/-30% DUOS component • Experiment 2: 9 cost levels, from +30% to -10% DUOS component • Experiment 3: 9 cost levels, from +30% to -10% DUOS component • However, for clarity of presentation, total bill sizes were presented
Stated Preference (SP) Survey Design • Example of choice from first experiment
Measuring Package Effects • Package Effect Problem: • respondents only ever see a subset of attributes in the lower-level experiments: never consider the entire set of services • may overestimate their willingness to pay for service improvements in the lower-level experiments • a respondent may say that they are prepared to pay 5p for 10 separate improvements but may be unwilling to pay 50p for all of them • Solution: • service valuations are produced for each attribute from the lower-level experiments • service valuations are also produced for each ‘package’ of variables reflecting the value if the package were improved to their best level • these are then compared to see whether we observe package effects
Stated Preference (SP) Survey Design • Example of choice from packaging experiment • Designed to reduce an over-statement of the total willingness to pay that can arise from all the lower level experiments
Characteristics of alternative/individual SP Methodology – Analysis • Analysis based on principles of utility maximisation • ‘Utility’ assigned to each alternative, made up of: • Attributes of the alternative • Characteristics of the respondent, e.g. income, etc. Random error reflecting influence of parameters not included, differences between individuals/responses Parameters to be estimated
SP Methodology – Analysis • Analysis based on principles of utility maximisation • ‘Utility’ assigned to each alternative, made up of: • Attributes of the alternative • Characteristics of the respondent, e.g. income, etc. • Assumption that error distribution is extreme value (i.i.d.) leads to logit model Assumptions: Random terms () are independent and identically distributed across alternatives They vary with an Extreme-Value distribution
SP Methodology – Analysis • Analysis based on principles of utility maximisation • ‘Utility’ assigned to each alternative, made up of: • Attributes of the alternative • Characteristics of the respondent, e.g. income, etc. • Assumption that error distribution is extreme value (i.i.d.) leads to logit model • Maximum likelihood techniques used to obtain values of unknown parameters • Find the βsthat optimise explanation of the model • Model outputs • Coefficient estimates (and t-statistics) • Model fit statistics
SP Methodology – Analysis • Coefficients used to: • Determine the relative importance of attributes • Determine the monetary value for attributes • Specify utility functions for prediction models • Monetary values for attributes identified in lower-level experiments • May be adjusted by values for groups of attributes identified by higher-level (packaging) experiment • This is a judgment call
Model Development • Analysis of data • Did respondents understand? • Did respondents trade between alternatives in the experiments? • Development of discrete choice models • Willingness-to-pay results
Respondents report high levels of understanding • After initial cleaning, 3193 data files with domestic and business customers
We also observe substantial trading within the Experiments LPN - Domestic LPN - Business Non-LPN - Domestic Non-LPN - Business
Respondents are sensitive to the cost levels being examined • Experiment 1: • Alternatives become unattractive at 10% increases • “As Now” just over 40% of choices chosen Domestic non-LPN Domestic LPN Business LPN Business non-LPN Change in bill size Change in bill size
Analysis of Costs – PACKAGING • Examining choices at different cost levels • Packaging: Alternatives become clearly unattractive high price increase for both domestic LPN and non-LPN cases In business case, the graphs are not so clear cut Domestic non-LPN Domestic LPN Business LPN Business non-LPN Change in bill size Change in bill size Packaging
Principles for Model Development • Data aggregated across DNOs, unless analysis indicates that customers from a specific DNO have statistically different valuations • Not the case for LPN, because of the different structure of the experiment, have therefore also used pilot data for LPN models • Continuous variables defined as differences from the current level, which is different for different DNOs • Have examined whether the willingness to pay for service improvements (per improvement) are different from the willingness to accept payment for service reductions (per reduction), • Have also tested bi-linear formulations, but none were justified
Model Development • Developed a joint model taking account of the differing error variations across 4 experiments • Testing for differences in valuation of various attributes by characteristics of the respondent and inertia on the “As Now”) Business: • DNO • business size • urban or rural locality • electricity usage level • industry type • region • whether company had experienced any power cuts • whether company had experienced an unplanned power cut • whether company had experienced a planned power cut Domestic: • DNO • urban or rural locality • respondent’s age • household income • electricity usage level • whether respondent had experienced any power cuts • whether respondent had experienced an unplanned power cut • whether respondent had experienced a planned power cut.
Model Development • Testing for best representation of service attributes • Categorical Represented linearly with differences for wtp and wta as appropriate • Testing for correlation between the “new” alternatives • Different model structures MNL (multinomial loglt) and NL (nested logit) tested for any evidence of correlation between 3 alternatives offered • NL model structure allows for different substitution patterns between alternatives Tests did not show a significant improvement in the mode fit with the NL model structure.The MNL structure is used in the subsequent analysis • Structural parameter • Structural parameter • Alternative B • As Now • Alternative A Nested logit structure
Model Development • Identifying and removing outliers from the data set • Respondents with extremely different responses to the rest of the samples • They can have a substantial impact on the model results • Analysis showed that there were some outliers in the non-LPN business model, which were dropped with a significant improvement in model fit • Corrections for repeated measures from the multiple data points collected from each respondents • SP allows us to collect several responses from each individual • The series of responses given are not independent • The “jack-knife” procedure used to correct for the interdependence of SP observations this may reduce t-ratios significantly • Testing for the existence of packaging effects • The lower level experiment can lead to an excessive value of total willingness to pay for all improvements (i.e., budgeting or halo effects) • Such an effect is observed and controlled for in the final valuations • Larger adjustments for domestic models • Larger adjustments for experiments 1 and 2, relative to 3
Model Findings – Willingness to Pay (wtp) • Packaging Adjustments • Effect is very prominent for Exp 1 and Exp 2 for residential customers • Also for LPN and Exp1 in non-LPN business customers
Model Results • For non-LPN models, have identified significant model parameters, with correct signs • Able to identify fewer parameters, with less significance, for LPN • LPN business model has very few significant parameters, even with pilot data • For domestic customers, we observe differing cost sensitivity by household income, which has a direct influence on wtp • For business customers, we observe differing cost sensitivity by size of business, which again impacts wtp
Power Cuts Description of levels 7 levels +1, +2, +3 Base (differs by DNO) -1, -2, -3 Unit: frequency of power cuts in 5 years or 10 years, depending on the DNO • Tested both wtp for reductions and wta deteriorations, around the DNO base value • Domestic non-LPN • significant differences between wtp and wta • single wtp term across DNOs; DNO-specific terms for wta • Domestic LPN • single coefficient for wtp and wta • Business, non-LPN • aggregated coefficients across DNOs, significant differences between wtp and wta • manufacturing businesses have higher wtp for reductions • different values observed for SP Manweb • Business, LPN • single coefficient for wtp and wta
Willingness-to-Pay: Frequency of Power Cuts(Domestic Customers)
Willingness-to-Pay (% of bill): Frequency of Power Cuts per 5 years(Business Customers)
Duration of Power Cuts Description of levels 9 levels +5, +10, +15,+20 Base (differs by DNO) -5, -10, -15, -20 Unit: average duration in minutes • Tested both wtp for reductions and wta deteriorations, around the DNO-value • Domestic non-LPN • significant differences between wtp and wta (except SSE Hydro, UU, SP Manweb) • generic improvement across all DNOs except ones mentioned above • generic deterioration across all DNOs except ones mentioned above and SSE-Southern, CE NEDL • Domestic LPN • single coefficient for wtp and wta • Business, non-LPN • aggregated coefficients across DNOs • significant differences between wtp and wta • Business, LPN • not able to identify a significant value
Willingness-to-pay: Change in average duration of power cuts(Domestic Customers)
Willingness-to-pay (% bill per minute): Change in average duration of power cuts(Business Customers)
Short Interruptions Description of levels 5 levels +1, +2 Base (differs by DNO) -1, -2 Unit: frequency of short interruptions in 5 years or 10 years, depending on the DNO • Tested both wtp for reductions and wta deteriorations, around the DNO-base value • Domestic non-LPN • significant differences between wtp and wta (except CE YEDL) • DNO-specific for CE YEDL, CE NEDL and SP Distribution, UU and SSE Hydro deteriorations,and EDF-EPN improvements • generic improvement across all DNOs except ones mentioned above • generic deterioration across all DNOs except ones mentioned above • Domestic LPN • single coefficient for wtp and wta • Business, non-LPN • aggregated coefficients across DNOs, except EDF-SPN • significant differences between wtp and wta • Business, LPN • single coefficient for wtp and wta, not significant
Willingness-to-pay: Number of short interruptions(Domestic Customers)
Willingness-to-pay (% bill per interruption): Number of short interruptions(Business Customers)
Information Description of levels 4 levels Automated messages or telephone operators to respond to customer calls (base) Base, plus call backs to provide information updates Base, plus text messages to provide information update Base, plus helpline for customers reliant on medical equipment (not in business survey) • Domestic non-LPN • generic value across all DNOs • Domestic LPN • single coefficient for wtp and wta • Business, non-LPN • Not able to identify any significant valuations • Business, LPN • Not able to identify any significant valuations
Willingness-to-pay: Provision of information (Domestic Customers)
Willingness-to-pay (% bill): Provision of information (Business Customers)
Restoration of Supply Description of levels 3 levels: Guarantee within 18 hours (base) Guarantee within 12 hours Guarantee within 6 hours • Valued very highly by domestic and business customers • Domestic non-LPN • Generic values for across a number of DNOs • DNO-specific values for EDF-EPN, SSE-Hydro, UU, WPD S. Wales, WPD S. West, SP Manweb • Domestic LPN • Observe higher values for those under 30 years of age (6 and 12 hour) and those with incomees over £60,000 (6 hour restoration) • Business, non-LPN • Significant valuations • Large companies value reduced restoration times more than small companies • Separate values for WPD South Wales • Business, LPN • Have identified valuations, insignificant after jack-knifing
Willingness-to-pay: Restoration of Supply(Domestic Customers)
Willingness-to-pay (% bill): Restoration of power supplies (Business Customers)
Compensation for failure to restore supply • Domestic non-LPN • Small value for fixed proportion, not possible to identify value for variable proportion • Domestic LPN • Values for fixed compensation identified • Business, non-LPN • Not able to identify any significant valuations • Business, LPN • Values for fixed compensation identified for LPN business customers, but not very significant Description of levels 4 levels of fixed compensation: £50, £100 (base), £150, £200 plus variable compensation for every additional 12 hour period, 4 levels: £25 (base), £50, £75, £100
Willingness-to-pay: Fix and Variable Compensation(Domestic Customers)
Willingness-to-pay (% bill per £): Compensation provided for failure to restore electricity in time (Business Customers) Fixed compensation Variable compensation
Multiple interruptions Description of levels 3 levels: Compensation after 5 interruptions Compensation after 4 interruptions (base) Compensation after 3 interruptions • Domestic non-LPN • small value identified for compensation after 3 interruptions rather than 4 interruptions, generic across all DNOs • Domestic LPN • small value identified for compensation after 3 interruptions rather than 4 interruptions
Willingness-to-pay: Compensation for Multiple Interruptions(Domestic Customers)
Notice for Planned interruptions Description of levels 3 levels: 2 days notice 5 days notice 10 days notice • Domestic non-LPN • generic across DNOs, differences by age bands • Domestic LPN • values obtained for changing notice period from 2 to 5 and 2 to 10 days – highest preference for 5 days notice • Business, non-LPN • Businesses who use from between 100kWh to 1MW+ of electricity, annually, valued an increase in the notice for planned interruptions • But we were not able to identify significant differences between 5 and 10 days notice periods • Business, LPN • We could not identify any values for the London businesses
Willingness-to-pay: Notice for planned interruptions(Domestic Customers) Notes: For non-LPN, the modelling results indicate that people younger than 50 prefer 5 days to 10 days; and people older than 50 don’t care at all. A higher preference for 5 days rather than 10 days is also shown in the LPN results.
Willingness-to-pay (% of bill): Advanced notice for planned interruptions(Business Customers)
Commitment to Underground Overhead lines • Domestic non-LPN • Generic across DNOs, but income effect, i.e. higher income households value this more highly, in addition to cost effect Description of levels 4 levels: None (base) 1.5% per annum 3% per annum 5% per annum
Willingness-to-pay: Undergrounding(Domestic Customers, non-LPN only)
Network resilience to storms Description of levels 5 levels +10%, +20% Base (differs by DNO) -10%, -20% Unit: number of customers affected • Tested both wtp for reductions and wta deteriorations, around the base value for each DNO • Domestic non-LPN • UU DNO has specific values, different wtp and wta • Otherwise, generic improvements across other DNOs • Generic deteriorations for DNOs, except UU, SSE Hydro, SP Distribution • Business, non-LPN • For business customers, we identified different valuations by DNO
Willingness-to-pay: Network Resilience to Storms(Domestic Customers)
Willingness-to-pay (£ per customer): Number of customers affected by major storm(Business Customers)