Session 19 25 October 2016 D Cale Reeves 1 and Varun Rai 12 1 LBJ School of Public Affairs The University of Texas at Austin 2 Department of Mechanical Engineering The University of Texas at ID: 565115
Download Presentation The PPT/PDF document "USAEE/IAEE, Tulsa" is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
Slide1
USAEE/IAEE, TulsaSession 19, 25 October 2016
D. Cale Reeves1 and Varun Rai1,2
1LBJ School of Public Affairs, The University of Texas at Austin2Department of Mechanical Engineering, The University of Texas at Austin
Behavioral Drivers of Solar PV Consumer “pull-forward” at Changes in Rebate LevelsSlide2
Solar PV Adopter Behavior at Rebate Changes :Overview
Motivations: CSI rebate stepdown designBackground: Two RD design concernsData: Recent survey Methods: Difference in means, logit modelResultsConclusionsImplications1Slide3
Solar PV Adopter Behavior at Rebate Changes:Motivations: CSI rebate
stepdown designCalifornia Solar Initiative disbursed >$2B in solar PV incentives (rebates) over ~10 yearsPolicy design: Installed capacity triggers “Stepdown” in rebate leveleg: 12/1/2008 PGE reaches 23.1MW, rebate drops from $1.90/W to $1.55/W Quasi-experiment lends itself to causal analysisRegression discontinuityUnique access to individual behavioral data – decision-making, information search processes, financial calculations, values – same time frame2Slide4
Solar PV Adopter Behavior at Rebate Changes:Background: RD design
Regression Discontinuity designs identify an effect by comparing values on opposite sides of a threshold in a forcing variableFigure adapted from Imbens and Lemieux (2007)discontinuous
3Slide5
Solar PV Adopter Behavior at Rebate Changes:Background: Two RD design concerns
First concern: Covariate ChangesOther variables shouldnot have discontinuityOther changes canconfound the effect
Common test
: look for discontinuity in covariatesIf NO: unlikely that covariates confoundIf YES: indication that the estimated effect MAY not be accurate
BUT
:
access to more covariates -> more robust testing
4Slide6
Solar PV Adopter Behavior at Rebate Changes:Background: Two RD design concerns
Second concern: “Pull-forward”Manipulation of the forcing variableIndividuals that choose a side self-select into treatment
C
ommon test: compare density on either side of thresholdIf NO: unlikely that individuals manipulate their status
If
YES
: indication that assignment was not “as-if-random”
BUT
:
imprecise manipulation -> not necessarily a problem
5Slide7
Solar PV Adopter Behavior at Rebate Changes:Data: Recent survey of solar adopters
Recent survey fielded to 6000 near-randomly selected California solar PV adopters in mid 2015156 variables, 7 sections including
Coverage across all rebate stepdown events, but thinLimited representation: cell-wise: .001-.03, total: .01 & .02Broader range of
covariates than typically available
System and decision details
Decision-making process
Sources of information
Financial aspects
690 responses (11.5%) – 194 relevant: 67 pre-, 127 post-
6
Stepdown
2 to 3
3 to 4
4 to 5
5 to 6
6 to 7
7 to 8
8 to 9
9 to 10
all
Pre –
8
6
2
1
8
8
14
20
67
Post –
12
12
6
12
25
18
21
21
127Slide8
Solar PV Adopter Behavior at Rebate Changes:Methods: Difference in means, logit model
Covariate changesDifference in means between pre-stepdown and post stepdown groupsFixed effect by rebate step compares pre- and post- groups across each stepdown event“Pull-forward”No density: Logit model on the DV pre-stepdown adopterIV: Indices from covariate changes analysis (savvy / unfocused)Fixed effect by month/quarter of the start of decision-making processSensitivity: reducing the bandwidth explores precision7
Slide9
Solar PV Adopter Behavior at Rebate Changes:Results: Covariate changes -> Difference in means
Pre-stepdown adopters are exhibiting more savvy consumer behaviorMore bids, more calculations, seek help to do calculations Post-stepdown adopters are less focused, less certain:Allow installers to initiate their decision making processValue broad, non-targeted information sourcesChoose installers because they offer monitoring, maintenance, etc.Dep Vars \ Ind Vars
Total number of bidsCalculated payback periodNeighbor helped w/ calculationsPre-stepdown0.701*0.127*
0.050**Observations194
194
194
Things pre-stepdown adopters do
more
than their post-stepdown counterparts:
Dep
Vars
\
Ind
Vars
Initiated by:
Direct Marketing
Important information:
Online tool
Important information:
Non-profit
Installer Choice:
Offer integrated product
Pre-stepdown
-0.112**
-0.441*
-0.322*
-0.087*
Observations
194
160
134
194
Things pre-stepdown adopters do
less than their post-stepdown counterparts:8Note: *p<0.1; **p<0.05; ***p<0.01Slide10
Solar PV Adopter Behavior at Rebate Changes:Results: “Pull-forward” -> logit models
Compared to adopters that start their decision making process at a similar time: More savvy adopters are more likely to adopt in a pre-stepdown window Less focused adopters are less likely to adopt in a pre-stepdown windowStronger and more consistent for less focused adoptersDependent variable: Pre-stepdownComparison within 70 day window
Compared to full sampleSavvy Index0.645*
0.5490.514**0.186Unfocused Index
-0.804**
-0.944***
-0.466**
-0.588***
FE: Initiation
Month
Yes
No
Yes
No
FE: Initiation
Quarter
No
Yes
No
Yes
9
Note: *p<0.1; **p<0.05; ***p<0.01Slide11
Solar PV Adopter Behavior at Rebate Changes:Results: “Pull-forward” -> logit models
Imprecise manipulation yields as-if-random-assignment when bandwidth is narrower than precision of manipulationConsistent results even as the bandwidth tightens Dependent variable: Pre-stepdown, compared to full sample by MonthBandwidth (each side)35 day30 day
25 day20 day15 day
10 day
5 day
Savvy Index
0.514**
0.335*
0.342
0.416*
0.490**
0.569*
0.474
Unfocused Index
-0.466**
-0.437**
-0.476**
-0.531**
-0.486*
-0.227
-0.367
Observations
552
552
552
552
552
552
552
Evidence of
manipulation
, but what of
precision
? (Similar models, abbreviated presentation)
10
Note: *p<0.1; **p<0.05; ***p<0.01Slide12
Solar PV Adopter Behavior at Rebate Changes:Results: “Pull-forward” -> logit models
Savvy adopters have relatively precise controlSimilar to other estimates: roughly 1 weekSuggests that behaviorally, Pre-stepdown adopters are not a great counterfactual for post-stepdown adoptersDependent variable: Pre-stepdown, compared within bandwidth by QuarterBandwidth
(each side)35 day30 day25 day
20 day
15 day
10 day
5 day
Savvy Index
0.549
0.712*
0.568
0.498
0.618
1.347**
2.071*
Unfocused Index
-0.944
***
-0.923
***
-0.924**
-0.844**
-0.750*
-0.505
-1.224
Observations
188
162
138
113
93
65
39
Evidence of
manipulation
, but what of
precision
? (Similar models, abbreviated presentation)
11
Note: *p<0.1; **p<0.05; ***p<0.01Slide13
Solar PV Adopter Behavior at Rebate Changes:Conclusions
Pre-stepdown adopters often have different decision-making processes than post-stepdown counter partsTheir information search is more focused, they get more bids, do more calculations, seek help when they need toSubset of adopters is likely able to precisely manipulate their treatment status Discontinuity incentivizes manipulationNot all adopters do (Savvy: 0=21%, 1=46%, 2=32%)12Slide14
Solar PV Adopter Behavior at Rebate Changes:Implications
Observed discontinuities + evidence of manipulation: Regression discontinuity analysis may estimate biased effectsSavvy adopters do have lower system pricesBUT: No mean difference in system price pre–/ post–Within groups, no savvy-ness OR within group savvy-ness constantPropensity weighting + RD may control for manipulation and improve estimates13Slide15
Solar PV Adopter Behavior at Rebate Changes: Closing
Thank youQuestions?References available upon requestD. Cale Reeves : d.cale.reeves@gmail.com14