/
Dynamic Learning ContextDependence in Sequential AttributeBased Stat Dynamic Learning ContextDependence in Sequential AttributeBased Stat

Dynamic Learning ContextDependence in Sequential AttributeBased Stat - PDF document

deborah
deborah . @deborah
Follow
343 views
Uploaded On 2021-10-07

Dynamic Learning ContextDependence in Sequential AttributeBased Stat - PPT Presentation

P Holmes and Kevin J Boyle presented that valuation response designed set mailback format discrimination between the sequence erence parameters shifts Lead a structural current choices positive corre ID: 897508

questions preference valuation parameter preference questions parameter valuation context parameters set alternatives lead bid test sequence stated choice significant

Share:

Link:

Embed:

Download Presentation from below link

Download Pdf The PPT/PDF document "Dynamic Learning ContextDependence in Se..." 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.


Presentation Transcript

1 Dynamic Learning Context-Dependence in S
Dynamic Learning Context-Dependence in Sequential, Attribute-Based, Stated-Preference Valuation Questions P. Holmes and Kevin J. Boyle presented that valuation response designed set mail-back format. discrimination between the sequence, erence parameters shifts. Lead a structural current choices positive correlation results indicate people learn their preferences levels across Contingent valuation (CV) been the commonly employed stated preference estimate nonmarket and Carson common response format, mous choice, valuation questions as specified cost (bid); respondents that amount The is typically A new stated-preference meth- ods has more been adopted for as attribute-based ABMs (Holmes and in various social science disciplines such as and marketing conceptual foundation for Land Economics February 2005 81 (1 ): 1 14-126 ISSN 0023-7639; E-ISSN 1543-8325 O models within economics finds its source demand for differentiated a total, value as CV, a set rel

2 evance (Adamowicz, Louviere, Blarney c
evance (Adamowicz, Louviere, Blarney contingent valuation, a use with preference ABMs (e.g., Holmes and most common format, the experiment," asks ferred alternative from a set con- taining two one or more attri- (e.g., Louviere, Hensher, and Swait description and a choice between alternative and merges contingent valuation and ABMs a hybrid stated-preference question. In order for a for researchers a sequence ence questions, pool responses withn and across when conducting design effi- 4,8, even 16 authors are, respectively, research forest econo- USDA Forest Service, and Libra Professor Environmental Economics, Resource Economics and The authors Loomis and Patty Champ, as anonymous referees, work. All the authors. 81 (1) Holmes Boyle: Stated- data for responses; (2) intra-individual structural components intra-individual responses maintained hypotheses Hicksian surplus a sequence has not ABM experiments fonnat. however, in ranking experiments Shirois

3 hi 1992), stated-preference and reveal
hi 1992), stated-preference and revealed-preference Louviere, and a relative parameter (for another) that can be lated from the preference parame- ters. Then, differences in scale, hypothesis be conducted evaluate whether more) parameter this proce- Ben-Akiva, Morikawa, and Shiroishi not stable a ranking experiment and recommended that unless preference parameters are experiment, the from a sequence second maintained hypothesis not depend on alterna- tives contained choices between not depend irrelevant alterna- important paper, Simonson and Tversky can be identified first context is "local" within a choice set. context within logit addition, they draw ground context," wherein the current alternatives present Although experiments identi- fylng background con- been reported and marketing research lit- eratures (Huber, Payne, and Tversky and 1993), in choice modeling. This oversight ths choice sets, tra-individual responses remain that are sampled population r

4 esponse behavior any particular indi- di
esponse behavior any particular indi- differ systematically from the average behavior due to factors, it model in permanent component idiosyncratic behavior and a transitory manent component shifts the indi- rect utility function, and distributed over probit model can be estimated for the permanent component probit ted-variable bias this study, question that and then frames the fonnat the status quo, "attribute-based referenda" a se- 82 (1) Holmes and Boyle: Valuation Questions parameter. As p - m (or, u - O), tic (Ben-Akiva and Leman 1985,72). people perfectly discriminate between j and the status the stan- E~ the other p --+ 0, not discriminate between alternative model predicts equal proba- between the alternatives = for YES over a se- learning about task and formulate responses, resulting discrimination between alternatives. Con- a decrease in p over a sequence indicates a as mental fatigue preference parame- ters, and scale can i

5 nto the learning, fatigue task complexit
nto the learning, fatigue task complexity emerging topic research (Johnson and Adamowicz 2001; DeShazo implied in [3], scale pa- preference parameters in multiplicative scale in any set. However, it an estimate than one data relative scale preference questions mong tives as they respond adjusted preference parameters for each valuation questions be isolated. adjusted parameters a likelihood preference parameters identical for each stage in a sequence data should not cases where the parame- ter stability not rejected, parameter for each the se- parameter for the ratio test. Identifying Context-Dependence Context independence a fundamental requirement for internal used in the devel- mental condition for internal known as states that an (x) from set a subset S T, must also C(T) function that from set statements were true: (1) {x} = C({x7y}); and (2) {y} = C({x7y, that the set cannot increase (the regularity property). Although Property can be between al

6 ternatives set (leading tests for mists)
ternatives set (leading tests for mists), social psychologists have global set alternatives under Tversky and Simonson distinguish between the set alternatives within the background context, containing alternatives have been pre- considered, and present a theoreti- accommodates both compo- as all valuation questions posed as binary 82 (1) tiolmes and Boyle: Stated- Valuation Questions {- 1*(bid, - bid,, ..)I lead_gain,+, = 0 if bidj,, � bidj,m = (1,2,3) otherwise 16bl if bid, � bid,+,,,m = (-1, -2, -3) otherwise [6c1 rid, bidj+, lead-loss, + , = 1 if bid, � bid,+,,m = (1,2,3). otherwise C6dI change in from reference Although a full speci- fication would include up lags and empirical specification only included two leads be- lags and leads were not statistically significant. lead variables enter the model specification di- as differenced variables. This s~ecification used because the non-price attributes are

7 each modeled as binary (011); clear,
each modeled as binary (011); clear, a priori, negative mar- ginal utility. considerations, indirect now be Note that, [7], change in the current the parameter estimate X. lag and U, with respect to bid, is the sum (h + xhiwdmt,,+m + ~~ie~d~a~poa,~+ni + ~AI~I~,,~ + n, + xiren-~(Jss,~ + nz), and the marginal a function the current alternative and the alternatives contained in compensating variation specific policy native package, the status quo, as the difference be- the inner product implicit prices (bk/-A) and attribute for the two pol- 2: and z: base and altered-policy scenarios. This specification assumes lag and gain and loss effects not present. prices contained in lag and lead choice have a statistically significant effect compensating variation relative com- marginal utility course, it lag and lead, gain and loss effects Identifjling Idiosyncratic Effcts Even after for dependence structural components be linked due to the per

8 manent, individual preferences. Kanni-
manent, individual preferences. Kanni- nen, and (1997) dom-effects model could be used Land Economics February 2005 TABLE 1 FOREST MANAGEMENT A~RIBUTES AND LEVELS Attribute Variable Name Levels Road density ROADS One road every mile n.a. One road % Post-harvest dead and n.a. Remove all DEAD5 Leave 5 DEAD10 per acre n.a. Remove all � 6" diameter LIVE153 153 per acre LIVE459 459 per acre Maximum size harvest n.a. 5 acres HOPEN35 35 acres HOPEN125 land available PERH80 80% harvest, 20% natural PERHSO 50% harvest, 50% natural n.a. harvest, 80% riparian protection H20ZONE500 At least n.a. Slash disposal n.a. where it Distribute along skid Bold letters common practice) indicate a benign practice, Hicksian surplus. CV questions. Unlike the double-bounded CV method, where commodity description commodity description changes Because each commodity description commodity descriptions they face completely randomized experi- mental

9 design), a posi- tive, statistically sig
design), a posi- tive, statistically significant p correlated sequence 111. The data for taken from a forest management study in the state The were struc- tured around a tract (23,000 scenarios (Table I), and attri- butes were coded dummy variables. payment vehicle used increase in taxes, and butes, and conversion a more environ- mentally benign management plan. in a booklet contained line drawings each attribute, provided descriptions tive and negative impacts level, and described the current forest agement conditions the study area. In the questionnaire, respondents their understanding the information the booklet, and were asked each attribute on a Likert familiarize respon- the attributes under consideration think carefully about the the preference initial sample were randomly sampled Maine drivers' licenses and Land fionomics February 2005 TABLE 3 TEST STATIS~CS FOR HYPOTHESES REGARDING SCALE PARAMETERS AND PREFERENCE PARAMETER STABILITY~ Test Q1

10 vs. Q2 Ql vs. Q3 Q1 vs. Q4 Relative
vs. Q2 Ql vs. Q3 Q1 vs. Q4 Relative scale lr, = 2,3,d& = I 1.05 0.825 1.23 Log-L, QI Q2, Q3, scaled, pooled LR test statistic (x2) 21.174 20.609 30.876 Reject Ho?: PI = 1 = P, = 2,3,4 LR test statistic (x2) 0.035 0.50 NIAc Reject Ho?: p, = , = IJ.~ = 23.4 NOd NOd N/Ac a on Swait x2 statistic for d.f. and 95% confidence ' Swait and method cannot test for identical hypothesis that preference parameters has been XZ statistic for 1 d.f. and 95% = habitat for and insects. Preference Parameter mates in Table 2 that the the fourth more informative than the models estimated Note parameters that the equation estimated from responses first question, third equation, and note that the pseudo- increased from the first equation fourth equation. Johnson and Des- vousges (1997), pair questions, concluded that later provide bet- preference than earlier Ada- mowicz (2001), ple-choice questions, butes were salient up about the b

11 ut that subsequent questions, respondent
ut that subsequent questions, respondents tended to on the brand name these results support point, but that there threshold where fatigue sets Results from and Louviere testing hypotheses preference pa- that a preferences occurred during tion sequence. valuation question that preference parameters changed for the the question for were most informative. These results responses for should not be pooled the prior questions, be- that the relative parameter for ques- than the preceding questions. Although and Louviere procedure to test whether this change statistically significant, the likelihood ratio test indicated erence parameters not identical, valuation question informative, and that re- spondents demonstrated discrimi- Holmes and Boyle: Stated- Preference kluation PARAMETER ESTIMATES MODELS wrr~ AND WITHOUT CONTEXT-DEPENDENT PARAMETERS Rescaled Q1-Q3 with Q4 with Rescaled Q1-Q3 Context-Dependence Context-Dependence -0.257*** -0.166 (0.124) -0.8

12 26""" (0.202) 0.155*** (0.062) 0.144""
26""" (0.202) 0.155*** (0.062) 0.144"" (0.064) 0.231 ** (0.107) DEAD5 0.187"" (0.082) 0.183"" (0.084) 0.396""" (0.131) DEAD10 0.197"" (0.079) 0.191"" (0.081) 0.249" (0.132) LIVE153 0.433""" (0.081) 0.449""" (0.083) 0.464""" (0.130) LIVE459 0.367""" (0.077) 0.367*** (0.079) 0.326""" (0.132) HOPEN35 0.008 (0.078) 0.255** (0.131) HOPEN125 -0.040 (0.078) - 0.033 (0.082) 0.254** (0.129) H20ZONE 0.028 (0.063) 0.032 (0.065) 0.121 (0.107) PERH80 -0.443""" (0.073) -0.473""" (0.076) -0.410""" (0.129) PERHSO -0.104 (0.076) -0.091 (0.077) -0.237"** (0.078) -0.245""" (0.081) -0.342*** (0.129) DSTSLASH 0.062 (0.076) 0.061 (0.079) -0.229" (0.129) h -0.00064*** (0.00029) -0.00074" (0.00045) ROADS,+, - -0.122"" (0.058) - LIVE153,+, - -0.181**" (0.064) - DEAD1O1- 1 - -0.170"" (0.078) - H20ZONEj- 1 - -0.149"" (0.071) - HOPEN1251-1 - - 0.254"" (0.1 13) A lead..lossj+ 1 - 0.000009 (0.00019) - 10,-loss, - 1 -

13 -0.00034* -0.000014 (0.00040) A iea
-0.00034* -0.000014 (0.00040) A ieaijarr,l+ 1 - 0.00027*** (0.00008) - A lag~atnj-1 - 0.00015" (0.000089) 0.00032** (0.0001 3) le(lil-/ots j+~ - 0.00006 (0.00027) - iag_/oss,l-2 - -0.0.000029 (0.00035) leod~nm,]+2 - 0.00016 (0.00010) - h iogjarn.,-2 - 0.00034*** (0.00013) 0.00035** (0.00015) o 0.104*** (0.035'1 0.1 08""" 10.036) - Note: standard errors in parentheses. *"* Significant at 0.01; ** significant at 0.05; * significant at nation between alternatives Dependence Rescaled parameter equal- ity could rejected (questions and panel models were estimated context- also es- timated for question The results context-depen- that attri- lag and This result for both and non-price attributes. parameter estimates for gains and (as defined [6a] through [6d]), prices had both direct and indirect (lead and lag bids) anticipated, results for the negative impact the current scenario decreased parameter estimates (i.e.,