Preference-Based Assessments| Volume 22, ISSUE 12, P1427-1440, December 01, 2019

# Carer Social Care-Related Quality of Life Outcomes: Estimating English Preference Weights for the Adult Social Care Outcomes Toolkit for Carers

Open AccessPublished:November 14, 2019

## Highlights

• There is increased interest in examining the effects of interventions on older people and their informal carers in economic evaluations. This study fills a gap in the literature by producing the first set of preference-based index values for the English version of Adult Social Care Outcomes Toolkit (ASCOT)-Carer in a general population sample from England using best-worst scaling.
• This work provides new evidence indicating attributes of social care-related quality of life of informal carers using the ASCOT-Carer are valued differently, highly valuing having time to do things valued and enjoyed, as well as having as much control over daily life as wanted.
• These preference weights for the English version of the ASCOT-Carer can be used for economic evaluation of interventions that support informal carers.

## Abstract

### Background

There is increasing interest in assessing the effects of interventions on older people, people with long-term conditions and their informal carers for use in economic evaluation. The Adult Social Care Outcomes Toolkit for Carers (ASCOT-Carer) is a measure that specifically assesses the impact of social care services on informal carers. To date, the ASCOT-Carer has not been preference-weighted.

### Objectives

To estimate preference-based index values for the English version of the ASCOT-Carer from the general population in England.

### Methods

The ASCOT-Carer consists of 7 domains, each reflecting aspects of social care-related quality of life in informal carers. Preferences for the ASCOT-Carer social care-related quality of life states were estimated using a best–worst scaling exercise in an online survey. The survey was administered to a sample of the general adult population in England (n = 1000). Participants were asked to put themselves into the hypothetical state of being an informal carer and indicate which attribute they thought was the best (first and second) and worst (first and second) from a profile list of 7 attributes reflecting the 7 domains, each ranging at a different level (1-4). Multinomial logit regression was used to analyze the data and estimate preference weights for the ASCOT-Carer measure.

### Results

The most valued aspect by English participants was the 'occupation' attribute at its highest level. Results further showed participants rated having no control over their daily life as the lowest attribute-level of all those presented. The position of the 7 attributes influenced participants’ best and worst choices, and there was evidence of both scale and taste heterogeneity on preferences.

### Conclusion

This study has established a set of preference-based index values for the ASCOT-Carer in England derived from the best–worst scaling exercise that can be used for economic evaluation of interventions on older individuals and their informal carers.

## Introduction

Informal care is a key part of the total care provided for older individuals and those with long-term conditions. Providing care for another individual can have significant effects on the carer’s health and quality of life (QoL).
• Al-Janabi H.
• Van Exel J.
• Brouwer W.
• Coast J.
A framework for including family health spillovers in economic evaluation.
• Al-Janabi H.
• Van Exel J.
• Brouwer W.
• et al.
Measuring health spillovers for economic evaluation: a case study in meningitis.
• Wittenberg E.
• Prosser L.A.
Disutility of illness for caregivers and families: a systematic review of the literature.
• Prosser L.A.
• Lamarand K.
• Gebremariam A.
• Wittenberg E.
Measuring family HRQoL spillover effects using direct health utility assessment.
Caregiving can have negative effects on the carer’s mental and physical health and QoL,
• Wittenberg E.
• Prosser L.A.
Disutility of illness for caregivers and families: a systematic review of the literature.
• Pinquart M.
• Sörensen S.
Differences between caregivers and noncaregivers in psychological health and physical health: a meta-analysis.
• Pinquart M.
• Sörensen S.
Correlates of physical health of informal caregivers: a meta-analysis.
but can also have positive effects, arising from empathy, altruism, and fulfillment.
• Al-Janabi H.
• Flynn T.N.
• Coast J.
Estimation of a preference-based carer experience scale.
• Basu A.
• Meltzer D.
Implications of spillover effects within the family for medical cost-effectiveness analysis.
Research shows interventions in older individuals and those with long-term conditions can have an effect on a carer’s health and QoL, so it is important that economic evaluations of these interventions also consider the impact on informal carers.
• Al-Janabi H.
• Flynn T.N.
• Coast J.
QALYs and carers.
• Goodrich K.
• Kaambwa B.
• Al-Janabi H.
The inclusion of informal care in applied economic evaluation: a review.
Several studies that have attempted to measure the effects of interventions on informal carers used the quality-adjusted life years (QALYs) metric.
• Al-Janabi H.
• Flynn T.N.
• Coast J.
QALYs and carers.
• Goodrich K.
• Kaambwa B.
• Al-Janabi H.
The inclusion of informal care in applied economic evaluation: a review.
In measuring QALYs of informal carers, the EuroQol-5D measure is often used.
• Goodrich K.
• Kaambwa B.
• Al-Janabi H.
The inclusion of informal care in applied economic evaluation: a review.
The EuroQol-5D focuses on aspects of health status and not on more holistic aspects of QoL or well-being, so it may not be broad enough to capture what matters to informal carers or the impact of caregiving on informal carers for economic evaluations.
• Reed C.
• Barrett A.
• Lebrec J.
• et al.
How useful is the EQ-5D in assessing the impact of caring for people with Alzheimer’s disease?.
To overcome this limitation, carer-specific measures assessing QoL or well-being have been developed.
• Michels C.T.
• Boulton M.
• Wee B.
• Peters M.
Psychometric properties of carer-reported outcome measures in palliative care: a systematic review.
• Harvey K.
• Catty J.
• Langman A.
• et al.
A review of instruments developed to measure outcomes for carers of people with mental health problems.
• Shilling V.
• Matthews L.
• Jenkins V.
• Fallowfield L.
Patient-reported outcome measures for cancer caregivers: a systematic review.
• Deeken J.F.
• Taylor K.L.
• Mangan P.
• Yabroff K.R.
• Ingham J.M.
Care for the caregivers: a review of self-report instruments developed to measure the burden, needs, and quality of life of informal caregivers.
Carer-specific measures typically focus on negative effects of care on carers’ QoL, such as care burden, while neglecting positive effects of care.
• Al-Janabi H.
• Flynn T.N.
• Coast J.
Estimation of a preference-based carer experience scale.
• Wiseman V.
Caring: the neglected health outcome? or input?.
Recent measures, such as the Care-related Quality of Life Instrument
• Brouwer W.B.F.
• van Exel N.J.A.
• van Gorp B.
• Redekop W.K.
The CarerQol instrument: a new instrument to measure care-related quality of life of informal caregivers for use in economic evaluations.
and the Carer Experience Scale,
• Al-Janabi H.
• Flynn T.N.
• Coast J.
Estimation of a preference-based carer experience scale.
• Al-Janabi H.
• Coast J.
• Flynn T.N.
What do people value when they provide unpaid care for an older person? A meta-ethnography with interview follow-up.
have been developed to capture the impact of caregiving on informal carers. However, validation is ongoing.
The Adult Social Care Outcomes Toolkit for Carers (ASCOT-Carer)
The ASCOT measure is disclosed in full herein but ordinarily should not be used for any purposes without the appropriate permissions of the ASCOT team and the copyright holder, the University of Kent. Please visit www.pssru.ac.uk/ascot or email [email protected] to inquire about permissions.
is an outcome measure aimed at assessing social care-related QoL (SCRQoL) for informal carers across 7 domains.
• Rand S.E.
• Malley J.N.
• Netten A.P.
• Forder J.E.
Factor structure and construct validity of the Adult Social Care Outcomes Toolkit for Carers (ASCOT-Carer).
• Rand S.E.
• Malley J.N.
• Netten A.P.
Measuring the Social Care Outcomes of Informal Carers: An Interim Technical Report for the Identifying the Impact of Social Care (IIASC) Study.
These include occupation, control over daily life, self-care, personal safety, social participation, space and time to be yourself, and feeling supported and encouraged. Each domain is rated on a 4-level scale, ranging from the ideal state (level 1) to high needs (level 4). A key feature of this measure, differentiating it from other relevant measures, is its focus on the impact of social care services on informal carers. The ASCOT-Carer was developed using interviews and cognitive testing to capture important aspects of carers’ SCRQoL.
• Rand S.E.
• Malley J.N.
• Netten A.P.
• Forder J.E.
Factor structure and construct validity of the Adult Social Care Outcomes Toolkit for Carers (ASCOT-Carer).
,
• Rand S.E.
• Malley J.N.
• Netten A.P.
Measuring the Social Care Outcomes of Informal Carers: An Interim Technical Report for the Identifying the Impact of Social Care (IIASC) Study.
,

Smith N, Holder J, Netten A. Measuring outcomes for carers. Paper presented at: British Society of Gerontology Annual Conference; September 4-6, 2008.

,
• Holder J.
• Smith N.
• Netten A.
Outcomes and Quality for Social Care Services for Carers: Kent County Council Carers Survey Development Project 2007-2008.
However, certain attributes of carers’ SCRQoL are likely to be more important than others. To account for this in producing a single overall SCRQoL score, we need to determine the relative value or weight for each of the measures’ attribute levels.
• Flynn T.N.
• Louviere J.J.
• Peters T.J.
• Coast J.
Best–worst scaling: what it can do for health care research and how to do it.
• Flynn T.N.
Valuing citizen and patient preferences in health: recent developments in three types of best–worst scaling.
To date, the ASCOT-Carer has not been preference weighted.
There are different methods available to elicit preferences in informal carers. Previous work has examined the use of the time-tradeoff technique for estimating the value of a carer’s well-being.
• Mohide E.A.
• Torrance G.W.
• Streiner D.L.
• Pringle D.M.
• Gilbert R.
Measuring the wellbeing of family caregivers using the time trade-off technique.
Additional work has estimated preference-based index values for the Carer Experience Scale using best–worst scaling (BWS) in carers.
• Al-Janabi H.
• Flynn T.N.
• Coast J.
Estimation of a preference-based carer experience scale.
BWS is arguably a less cognitively burdensome method compared to other choice methods, such as discrete choice experiments or time-tradeoff technique.
• Whitty J.A.
• Walker R.
• Golenko X.
• Ratcliffe J.
A think aloud study comparing the validity and acceptability of discrete choice and best worst scaling methods.
• Cheung K.L.
• Wijnen B.F.M.
• Hollin I.L.
• et al.
Using best-worst scaling to investigate preferences in health care.
The main advantage of the BWS (profile case) is people consider the attribute-levels that describe a profile, instead of comparing 2 profiles
• Cheung K.L.
• Wijnen B.F.M.
• Hollin I.L.
• et al.
Using best-worst scaling to investigate preferences in health care.
• Flynn T.N.
• Louviere J.J.
• Peters T.J.
• Coast J.
Using discrete choice experiments to understand preferences for quality of life. Variance-scale heterogeneity matters.
(for further details on this method, see Flynn and Marley
• Flynn T.N.
• Marley A.A.J.
Best-worst scaling: theory and methods.
).
The aim of this study was to estimate a set of preference weights for the English version of the ASCOT-Carer instrument gathered from an English sample using BWS. This paper begins by explaining the methods used for the BWS exercise: experimental design, methods of data collection, and planned analysis. We report BWS model results relating to the generation of preference weights for the ASCOT-Carer in an English sample for economic evaluations.

## Methods

### Valuation Exercise

During BWS, participants were asked to put themselves in the imaginary state of caring for someone who was unable to care for themselves owing to illness, accident, or old age. Participants were presented with 8 scenarios. Each scenario included a profile containing 7 attributes reflecting the 7 SCRQoL domains of ASCOT-Carer.
• Flynn T.N.
• Louviere J.J.
• Peters T.J.
• Coast J.
Best–worst scaling: what it can do for health care research and how to do it.
• Flynn T.N.
• Louviere J.J.
• Peters T.J.
• Coast J.
Using discrete choice experiments to understand preferences for quality of life. Variance-scale heterogeneity matters.
• Flynn T.N.
• Marley A.A.J.
Best-worst scaling: theory and methods.
The attributes each represented 1 of 4 levels, ranging from ideal state (level 1) to high needs (level 4). Participants were asked to select the best choice from the list of attribute-levels in the scenario (type of BWS experiment known as 'profile case'). This selected choice was grayed out. The same process was repeated for the worst, second best, and second worst choices. After selecting all 4 choices in the first scenario, this process was repeated for the remaining 7 scenarios of the different attribute-level combinations. In total, each participant made 32 choices (ie, 4 choices in each of the 8 scenarios). (See Appendix Figure 1 in Supplemental Materials found at https://doi.org/10.1016/j.jval.2019.07.014 for an example of the BWS task using the ASCOT-Carer measure).

### Experimental Design

The scenarios for the BWS exercise were developed using an orthogonal main effects plan.
• Louviere J.J.
• Flynn T.N.
• Marley A.A.J.
Best-Worst Scaling: Theory, Methods and Applications.
All attributes had the same number of levels, so we were able to use a balanced orthogonal main effects plan whereby all attributes were statistically independent of one another. The full factorial design plan consisted of 47 possible profiles, which would be too many states for presentation.
• Flynn T.N.
Valuing citizen and patient preferences in health: recent developments in three types of best–worst scaling.
• Flynn T.N.
• Huynh E.
• Peters T.J.
• et al.
Scoring the ICECAP-A capability instrument. Estimation of a UK general population tariff.
• Ratcliffe J.
• Huynh E.
• Stevens K.
• Brazier J.E.
• Sawyer M.
• Flynn T.
Nothing about us without us? A comparison of adolescent and adult health-state values for the child health utility-9D using profile case best-worst scaling.
The fractional-factorial design reduced the full factorial plan to a design matrix of 32 scenarios. The design matrix was blocked into 4 segments. Thus, each participant received 8 BWS scenarios. The blocking procedure retained balance within the blocks and sought to minimize correlations of the levels being presented for the attributes within the block. A foldover design was used to eliminate easy or straightforward choices from each scenario.
• Johnson F.R.
• Lancsar E.
• Marshall D.
• et al.
Constructing experimental designs for discrete-choice experiments: report of the ISPOR conjoint analysis experimental design good research practices task force.
The blocked scenarios were randomly allocated to participants to minimize selection bias. The order of attributes was randomized between participants to prevent ordering bias and separate the effect of attribute choice from the position of that attribute within a scenario.
• Day B.
• Bateman I.J.
• Carson R.T.
• et al.
Ordering effects and choice set awareness in repeat-response stated preference studies.
• Campbell D.
• Erdem S.
Position bias in best-worst scaling surveys: a case study on trust in institutions.

### Survey Design and Sampling

The BWS exercise was part of a self-completion online survey. The survey included some general sociodemographic questions to assess representativeness of the sample and participants’ consent to take part in the study. The survey also included a set of questions regarding participants’ QoL and SCRQoL using the ASCOT-Carer measure, the BWS exercise, follow-up questions about participants’ understanding of the BWS exercise, questions concerning the participants’ experiences of social care and caring, and some additional sociodemographic and socioeconomic questions. The study was reviewed and approved by the University of Kent SRC Research Ethics Committee, [REF SR CEA 149].
We estimated preferences of a general population sample (rather than a service user or carer sample), which is a common approach in the estimation of preference weights.
• Netten A.
• Burge P.
• Malley J.
• et al.
Outcomes of social care for adults: developing a preference-weighted measure.
• Dolders M.G.T.
• Zeegers M.P.A.
• Groot W.
• Ament A.
A meta-analysis demonstrates no significant differences between patient and population preferences.
This is because it is the public at large whose views are relevant, where services are publicly-funded, and whose data are used to make decisions about resource allocations.
• Dolan P.
Whose preferences count?.
The survey was piloted in May 2016 with a total sample of 50 adults from the general population, recruited through an online panel. The pilot data helped inform decisions regarding wording of the BWS exercise. After the pilot, some questionnaire items and wording of the BWS exercise were refined for clarity.
The main survey was conducted between June and July 2016. The study included 1000 adults recruited from the general population in England. Participants were recruited from the same online panel as the pilot; those who completed the main survey did not complete the pilot survey. Sampling was targeted to be representative of the English general population in age, sex, and region. Individuals who took less than 4.5 minutes to complete the BWS task were omitted from the sample before the end of data collection, as this was deemed an unrealistically short period of time to complete the task. Sampling continued until the target of 1000 participants was reached. No further exclusion criteria were applied for the analysis.

### Statistical Analysis

#### Analysis of best-worst scaling data

Based on random utility theory, a multinomial logit regression (MNL)
Conditional logit analysis of qualitative choice behavior.
• Train K.
Discrete Choice Methods in Simulation.
model was used to estimate preference weights for informal carers’ SCRQoL using the ASCOT-Carer. The estimation closely followed Netten et al.
• Netten A.
• Burge P.
• Malley J.
• et al.
Outcomes of social care for adults: developing a preference-weighted measure.
Each attribute was specified as an alternative and given a utility function, which was based on the level at which the attribute was presented within the scenario and the position of the attribute in the scenario. The position effect of the attribute was separated by best (first and second) and worst (first and second) choices. The MNL model assumed all choices were independent and sequential.
• Flynn T.N.
• Louviere J.J.
• Peters T.J.
• Coast J.
Using discrete choice experiments to understand preferences for quality of life. Variance-scale heterogeneity matters.
• Huynh E.
• Coast J.
• Rose J.
• Kinghorn P.
• Flynn T.
Values for the ICECAP-Supportive Care Measure (ICECAP-SCM) for use in economic evaluation at end of life.
The basic MNL model was estimated as follows:
$Equation 1.$
(1)

where Uiq is the utility function for respondent q derived for an alternative i being chosen from a profile of J alternatives. The utility function has a systematic component, $Viq$, and a random component, $εiq$.
An example of the specification of the basic utility function for the occupation domain is outlined below. Effects coding was used to dissociate best (first and second) and worst (first and second) choices:
$Uq(occupation)=+β1∗(1,ifoccupationlevel=1)iq∗(1,ifchoice=bestorsecondbest)iq$

$−β1∗(1,ifoccupationlevel=1)iq∗(1,ifchoice=worstorsecondworst)iq$

$⋮$

$+β4∗1,ifoccupationlevel=4iq∗1,ifchoice=bestorsecondbestiq$

$−β4∗1,ifoccupationlevel=4iq∗1,ifchoice=worstorsecondworstiq$

$+γ1B∗1,ifoccupationappearedinfirstrowiq∗1,ifchoice=bestorsecondbestiq$

$⋮$

$+γ7B∗1,ifoccupationappearedinseventhrowiq∗1,ifchoice=bestorsecondbestiq$

$−δ1W∗(1,ifoccupationappearedinfirstrow)iq∗(1,ifchoice=worstorsecondworst)iq$

$⋮$

$−δ7W∗1,ifoccupationappearedinseventhrowiq∗1,ifchoice=worstorsecondworstiq+εiq$
(2)

where $β1…β4$denotes the coefficient for each attribute-level (1 refers to the ideal state, whereas 4 refers to high needs), () are the coefficients for the position of the occupation attribute within the best–worst scenario if the choice was best or second best (worst or second worst), $εiq$ refers to the random component. The attribute of control over daily life at level 4 was used as a reference level and was set to 0.
• Netten A.
• Burge P.
• Malley J.
• et al.
Outcomes of social care for adults: developing a preference-weighted measure.
Each choice in the above was estimated using the MNL model:
$Equation 3.$
(3)

where $Piq$ refers to the probability of each respondent q choosing alternative i from all relevant alternatives j in a profile J. $ϑ$ is the scale parameter and inversely proportionate to the standard deviation of the random component.
The basic MNL model was first estimated. The basic model refers to the MNL model (1) in which $ϑ$ = 1. T-ratios were used to indicate the level of significance of the coefficient compared to the reference levels (control over daily life at level 4 and first position of the profile list for first and second best and worst choices). A t-ratio of -1.96 or +1.96 was considered statistically significant at the 95% level.
The scale heterogeneity MNL (S-MNL) model
• Fiebig D.G.
• Keane M.P.
• Louviere J.
• Wasi N.
The generalized multinomial logit model: accounting for scale and coefficient heterogeneity.
was estimated to control for differences in error variance in subgroups. This allowed us to investigate the consistency of choices and would allow for more valid and reliable utility estimates.
• Flynn T.N.
• Louviere J.J.
• Peters T.J.
• Coast J.
Using discrete choice experiments to understand preferences for quality of life. Variance-scale heterogeneity matters.
Following the work of Netten et al,
• Netten A.
• Burge P.
• Malley J.
• et al.
Outcomes of social care for adults: developing a preference-weighted measure.
scale factors were included in the model to test for scale heterogeneity based on previous research (age, education level, best and worst choices, time taken to complete the BWS task, health status).
• Netten A.
• Burge P.
• Malley J.
• et al.
Outcomes of social care for adults: developing a preference-weighted measure.
After testing the hypothesized scale factors, we ended with a model that included 3 statistically significant scale factors that were sensible and in-line with economic/psychological theory: age, education, and time taken to complete the BWS task.
It is also important to control for variation in preferences between subgroups associated with observable characteristics (taste heterogeneity). Our aim was to account for any additional variation in estimation within the model based on our sampling approach. Taste heterogeneity was modeled by adding interaction terms between attribute levels and observable characteristics to the systematic component of the model (1).
• Louviere J.J.
• Flynn T.N.
• Marley A.A.J.
Best-Worst Scaling: Theory, Methods and Applications.
Several taste factors were included in the model to test for taste heterogeneity on the attribute levels based on socioeconomic and sociodemographic characteristics that were either significantly under- or overrepresented in the sample compared to the general population (see Table 1).
• Netten A.
• Burge P.
• Malley J.
• et al.
Outcomes of social care for adults: developing a preference-weighted measure.
After testing the hypothesized taste factors, we ended with a model that included 4 statistically significant taste factors that were sensible and in-line with economic/psychological theory: education, marital status, social grade, and religion.
Table 1Sample characteristics compared to the general population (N = 1000).
VariableSampleGeneral population
frequency%frequency%
Sex
Office for National Statistics
2011 Census for England, 18+ population from England and Wales.
Male48048.0020 262 82248.62
Female52052.0021 412 67451.38
Age
Office for National Statistics
Analysis Tool.
(years)
18-2410610.604 920 12811.41
25-3417517.507 485 99617.37
35-4416716.707 107 37216.49
45-5418218.207 700 36017.86
55-6419019.006 183 04314.34
65-7917317.307 089 98316.45
80+70.702 621 5896.08
Ethnicity
Office for National Statistics
2011 Census for England, 18+ population from England and Wales.
White90090.0036 377 82987.29
Mixed/multiple ethnic backgrounds151.50602 8621.45
Asian/Asian British626.203 007 1107.22
Black/African/Caribbean/Black151.501 284 2813.08
Other ethnic group10.10403 4140.97
Prefer not to say70.70----
Religion
Office for National Statistics
2011 Census for England, 18+ population from England and Wales.
No religion42642.609 768 62223.44
Christian (all denominations)48048.0025 721 73561.72
Buddhist/Hindu/Jewish/Muslim646.403 063 8747.35
Any other religion/prefer not to say303.003 121 2657.49
Education (ISCED class)
Office for National Statistics
Census for England, population estimates for those aged 16+ from England and Wales.
Below secondary education (ISCED <2)494.9015 371 25135.76
Lower secondary education and upper secondary education (ISCED 2, 3)40640.606 544 61415.22
Short-cycle tertiary and post-secondary education (ISCED 4, 5)14014.006 842 56515.92
BA/MA/PhD or equivalent (ISCED 6, 7, 8)38938.9011 769 36127.38
Don’t know30.30----
Other161.602 461 8295.73
Marital status
• Wittenberg E.
• Prosser L.A.
Disutility of illness for caregivers and families: a systematic review of the literature.
Married/in a civil partnership58258.2020 129 65746.82
Separated (still legally married)202.001 141 1962.65
Divorced575.703 857 1378.97
Widowed313.102 971 7026.91
Single, that is, never married and never in a civil partnership29929.9014 889 92834.64
Prefer not to say111.10----
Employment status
Office for National Statistics
2011 Census for England, population estimated for those aged between 16 and 74 from England and Wales.
Employed (full-time, part-time, self-employed)61661.6024 143 46462.10
In education (not paid for by employer), even if on vacation383.803 592 6549.24
Unemployed424.201 702 8474.38
Permanently sick or disabled242.401 574 1344.05
Retired22422.405 320 69113.68
In community or military service00.00----
Doing housework, looking after children or other persons505.001 695 1344.36
Other30.30852 4502.19
Don’t know30.30----
2011 Census for England, all residents for England between 16 and 64.
A/B46246.207 737 60222.94
C127627.6010 238 03930.35
C212212.207 396 56921.93
D/E13013.008 362 13824.79
Other101.00----
Self-reported health status
Office for National Statistics
Census for England, all residents from England and Wales.
Very good19619.6025 005 71247.17
Good50250.2018 141 45734.22
Fair24324.306 954 09213.12
BA indicates bachelor of arts; MA, master of arts; ISCED, International Standard Classification of Education ; PhD, doctor of philosophy.
Each participant in the BWS task made 32 choices (giving 32 000 observations in total). The full sample was used for the basic MNL model but was reduced to 31 392 observations due to a small number of missing observations for the education variable (608 observations, or 19 participants) in the S-MNL and taste heterogeneity S-MNL models.
The MNL models were developed and estimated first using ALOGIT software. To correct for the repeated nature of the task, robust standard errors were obtained using the sandwich estimator
• Netten A.
• Burge P.
• Malley J.
• et al.
Outcomes of social care for adults: developing a preference-weighted measure.
and estimated in Biogeme.

Bierlaire M. BIOGEME: a free package for the estimation of discrete choice models. Paper presented at: 3rd Swiss Transport Research Conference; 2003; Ascona, Switzerland.

#### Generation of preference-based index values for the ASCOT-Carer in England

Results from the taste heterogeneity S-MNL model were then used to generate sensible preference-based index values by taking into account representativeness of the data. Population proportions were applied to certain coefficients where there was evidence of (taste) variation in the taste heterogeneity S-MNL model to produce revised average values that take into account significant differences that exist between groups. Socioeconomic and sociodemographic variables selected to apply population proportions were informed by large differences compared to the general population detected through descriptive statistics (>10 point difference): marital status, education, social grade, and religion.
• Huynh E.
• Coast J.
• Rose J.
• Kinghorn P.
• Flynn T.
Values for the ICECAP-Supportive Care Measure (ICECAP-SCM) for use in economic evaluation at end of life.

Burge P, Potoglou D, Kim CW, Hess S. How do the public value different outcomes of social care? Estimation of preference weights for ASCOT. 2010.

Population proportions taken from English Census and national population statistics were applied to the selected taste variables to better reflect tastes of the English population (see Appendix Table 1 in Supplemental Materials found at https://doi.org/10.1016/j.jval.2019.07.014 for list of sources).
The next step was to rescale the revised average values so that summed state attribute scores varied on a 0-1 interval.
• Flynn T.N.
• Huynh E.
• Peters T.J.
• et al.
Scoring the ICECAP-A capability instrument. Estimation of a UK general population tariff.
• Huynh E.
• Coast J.
• Rose J.
• Kinghorn P.
• Flynn T.
Values for the ICECAP-Supportive Care Measure (ICECAP-SCM) for use in economic evaluation at end of life.
• Coast J.
• Flynn T.N.
• Natarajan L.
• et al.
Valuing the ICECAP capability index for older people.
In the QALY, death is typically anchored to 0. For the current study, values were rescaled such that high needs or the pits state were given a value of 0. This reflected a measure of unmet needs. The benefit of this option is it allowed us to measure and understand the relative value of each SCRQoL state and retain a scale of high needs or pits state and ideal state.
• Flynn T.N.
• Huynh E.
• Peters T.J.
• et al.
Scoring the ICECAP-A capability instrument. Estimation of a UK general population tariff.
• Huynh E.
• Coast J.
• Rose J.
• Kinghorn P.
• Flynn T.
Values for the ICECAP-Supportive Care Measure (ICECAP-SCM) for use in economic evaluation at end of life.
• Coast J.
• Flynn T.N.
• Natarajan L.
• et al.
Valuing the ICECAP capability index for older people.
Anchoring values of the ASCOT-Carer to the dead state may be considered in future work.
One-seventh of the value of the state 4444444 (high needs for all 7 domains) was subtracted from all attributes. This value was then divided by the difference between value states 1111111 (ideal state for all 7 domains) and 4444444 (high needs for all 7 domains). This was to ensure the lowest possible state (pits state) (high needs for all 7 domains) sums to 0 and the highest possible state (ideal state for all 7 domains) sums to 1, while maintaining relative differences between the attribute-level coefficients.
• Flynn T.N.
• Huynh E.
• Peters T.J.
• et al.
Scoring the ICECAP-A capability instrument. Estimation of a UK general population tariff.
• Huynh E.
• Coast J.
• Rose J.
• Kinghorn P.
• Flynn T.
Values for the ICECAP-Supportive Care Measure (ICECAP-SCM) for use in economic evaluation at end of life.
• Coast J.
• Flynn T.N.
• Natarajan L.
• et al.
Valuing the ICECAP capability index for older people.

## Results

Table 1 includes sociodemographic and socioeconomic statistics from the sample (n = 1000) compared to the general population.
Office for National Statistics
2011 Census for England, 18+ population from England and Wales.
Office for National Statistics
Analysis Tool.
Office for National Statistics
Census for England, population estimates for those aged 16+ from England and Wales.
Office for National Statistics
2011 Census for England, population estimated for those aged between 16 and 74 from England and Wales.
2011 Census for England, all residents for England between 16 and 64.
Office for National Statistics
Census for England, all residents from England and Wales.
There was a larger number of respondents in higher education (eg, having a degree and above [bachelor of arts, master of arts, doctor of philosophy, or equivalent]) (38.9%) compared to the general population (27.4%). There was a larger number of respondents in either lower or upper secondary education (40.6%) and a smaller proportion of respondents below secondary education (4.9%) compared to the general population (15.2% and 35.8%). For religion, there was a larger proportion of respondents who reported no religion (42.6%) compared to the general population (23.4%). The sample also underrepresented those of Christian faith. There was a larger proportion of respondents reported as married/in a civil partnership (58.2%) compared to the general population (46.8%). For social grade, grades AB (combined) were overrepresented (46.2%) compared to the general population (22.9%). Both grade C2 and grades DE (combined) were also underrepresented. The percentage of people with self-reported very good health status also differed. Good and fair health statuses (50.2% and 24.3%) were also overrepresented compared to the general population (34.2% and 13.1%).
Descriptives for the BWS task are reported in Table 2. The median time to complete the BWS task was just under 8 minutes. Nearly all of the participants were able to put themselves into the imaginary situations, either all of the time (45.8%) or some of the time (47.6%). Interestingly, 23.5% of the sample did not think about the length of time in the imaginary situations, whereas 31.4% of participants imagined the length of time in the imaginary situations would be a number of years, and 23.4% thought the length of time would be permanent. Most respondents were able to understand the scenarios presented in the BWS task (73.4%), considered all of the scenarios when making decisions (82.7%), and found the BWS task fairly easy to complete (63.5%).
Table 2Descriptives for the BWS task (N = 1000).
Variable%
Time taken to complete the BWS task (minutes)—median IQR: 25%-75%7.88
6.00-11.37
Ability to put themselves in the imaginary situation
Yes, all of the time45.80
Yes, but only some of the time47.60
No6.60
Assumed length of time in imaginary situation
Temporary—less than a few weeks2.10
Temporary—a number of weeks3.80
Temporary—a number of months7.00
Temporary—a number of years31.40
Permanent or rest of my life23.40
Understanding of the scenarios in the BWS task
Yes, all of the time73.40
Yes, but only some of the time24.00
No2.60
Whether considered all of the scenarios when making decisions
Yes, all of the time82.70
Yes, but only some of the time15.80
No1.50
Report of how easy or difficult to complete the BWS task
Very easy12.10
Fairly easy63.50
Fairly difficult22.80
Very difficult1.60
BWS indicates best-worst scaling; IQR, interquartile range.

### Best–Worst Scaling Model Results

Table 3 presents the attribute-level coefficients, the position best (first and second) and worst (first and second) choice variables estimated using the basic MNL model for the ASCOT-Carer. The goodness-of-fit measure (rho-squared) indicated the MNL model fit performed relatively well, with a rho-squared value = 0.226 (a rho-squared value between 0.25 and 0.3 is equivalent to a value between 0.75 and 0.80 of a linear regression model).
• Netten A.
• Burge P.
• Malley J.
• et al.
Outcomes of social care for adults: developing a preference-weighted measure.
Table 3Estimated parameters for the ASCOT-Carer measure using general population data from England-Basic MNL model (N = 1000).
Attribute-levelMNL
CoefficientSEt -ratio (robust)
Occupation
• 1.
I’m able to spend my time as I want, doing things I value or enjoy.
4.0190.11235.9
• 2.
I’m able do enough of the things I value or enjoy with my time.
3.7480.10535.6
• 3.
I do some of the things I value or enjoy with my time, but not enough.
2.1700.07329.6
• 4.
I don’t do anything I value or enjoy with my time.
0.3270.0496.6
Control over daily life
• 1.
I have as much control over my daily life as I want.
3.8760.10935.6
• 2.
I have adequate control over my daily life.
3.2880.09634.3
• 3.
I have some control over my daily life, but not enough.
1.8250.06926.3
• 4.
I have no control over my daily life.
0.0000.000Constant
Looking after yourself
• 1.
I look after myself as well as I want.
3.1220.08935.2
• 2.
I look after myself well enough.
2.9570.08933.3
• 3.
Sometimes I can’t look after myself well enough.
0.8390.05515.2
• 4.
I feel I am neglecting myself.
0.4510.0538.6
Safety
• 1.
I feel as safe as I want.
2.9430.08235.7
• 2.
Generally I feel adequately safe, but not as safe as I would like.
1.7700.06328.3
• 3.
I feel less than adequately safe.
1.0660.05718.6
• 4.
I don’t feel at all safe.
0.6010.05411.1
Social participation and involvement
• 1.
I have as much social contact as I want with people I like.
3.0950.09333.4
• 2.
I have adequate social contact with people.
2.7800.08134.4
• 3.
I have some social contact with people, but not enough.
1.8940.06628.6
• 4.
I have little social contact with people and feel socially isolated.
0.7760.05414.4
Space and time to be yourself
• 1.
I have all the space and time I need to be myself.
3.6810.10335.8
• 2.
I have adequate space and time to be myself.
3.2940.09235.7
• 3.
I have some of the space and time I need to be myself, but not enough.
2.0080.07028.6
• 4.
I don’t have any space or time to be myself.
0.5170.04910.6
Feeling supported and encouraged
• 1.
I feel I have the encouragement and support I want.
3.2550.09335.1
• 2.
I feel I have adequate encouragement and support.
3.0740.08835.0
• 3.
I feel I have some encouragement and support, but not enough.
1.8580.06727.8
• 4.
I feel I have no encouragement and support.
0.6520.05412.1
Attribute position in the BWS task
Position 1_B0.0000.000Constant
Position 2_B−0.1400.041−3.4
Position 3_B−0.2220.041−5.4
Position 4_B−0.3140.042−7.5
Position 5_B−0.3650.043−8.4
Position 6_B−0.4020.045−9.0
Position 7_B−0.3970.045−8.8
Position 1_W0.0000.000Constant
Position 2_W0.0100.0430.2
Position 3_W−0.0080.043−0.2
Position 4_W0.0230.0450.5
Position 5_W0.0340.0440.8
Position 6_W−0.0220.045−0.5
Position 7_W0.0620.0451.4
No. of observations32 000
df39
Final log-likelihood−41 693.4
Rho2 (0)0.226
AIC83 464.9
BIC83 791.5
AIC indicates Akaike information criterion; ASCOT, Adult Social Care Outcomes Toolkit; BIC, Bayesian information criterion; BWS indicates best-worst scaling; MNL, multinomial logit regression; SE, standard error.
All attribute-levels were estimated relative to level 4 of the 'control over daily life' attribute. The latter was defined as the reference level because it had the lowest utility. There were statistically significant differences between estimated coefficients compared to level 4 of the 'control over daily life' attribute. This indicates all other SCRQoL states were more valued compared to 'control over daily life' level 4, indicating these weights were greater and more positive compared to this attribute-level.
The largest coefficient was estimated for the 'occupation' attribute at level 1. The second largest coefficient was the 'control over daily life' attribute at level 1. The lowest coefficient was estimated for the 'control over daily life attribute' at level 4, followed by the second-lowest coefficient, 'occupation' attribute at level 4.
The parameters of the position variables capturing instances in which an attribute was chosen in a particular position for the best and second-best choices were all statistically significant. There was no clear trend for positioning of worst and second-worst choices.
A second model was estimated to investigate the potential influence of scaling effects on preferences. Results for the S-MNL model with scale effects are presented in Table 4.
Table 4Estimated parameters for the ASCOT-Carer measure using general population data from England S-MNL model (N = 981).
Attribute-levelS-MNL
CoefficientSEt ratio (robust)
Occupation
• 1.
I’m able to spend my time as I want, doing things I value or enjoy.
2.8860.18915.3
• 2.
I’m able do enough of the things I value or enjoy with my time.
2.6940.17615.3
• 3.
I do some of the things I value or enjoy with my time, but not enough.
1.5610.10614.7
• 4.
I don’t do anything I value or enjoy with my time.
0.2330.0376.2
Control over daily life
• 1.
I have as much control over my daily life as I want.
2.7810.18515.1
• 2.
I have adequate control over my daily life.
2.3680.15515.3
• 3.
I have some control over my daily life, but not enough.
1.3040.09214.2
• 4.
I have no control over my daily life.
0.0000.000Constant
Looking after yourself
• 1.
I look after myself as well as I want.
2.2280.15114.7
• 2.
I look after myself well enough.
2.1130.14514.6
• 3.
Sometimes I can’t look after myself well enough.
0.6100.05211.8
• 4.
I feel I am neglecting myself.
0.3420.0418.4
Safety
• 1.
I feel as safe as I want.
2.0920.14214.7
• 2.
Generally I feel adequately safe, but not as safe as I would like.
1.2780.08714.7
• 3.
I feel less than adequately safe.
0.7790.05913.2
• 4.
I don’t feel at all safe.
0.4470.04510.0
Social participation and involvement
• 1.
I have as much social contact as I want with people I like.
2.2210.14615.2
• 2.
I have adequate social contact with people.
2.0080.13215.2
• 3.
I have some social contact with people, but not enough.
1.3550.09514.2
• 4.
I have little social contact with people and feel socially isolated.
0.5450.05010.8
Space and time to be yourself
• 1.
I have all the space and time I need to be myself.
2.6400.17515.1
• 2.
I have adequate space and time to be myself.
2.3650.15515.3
• 3.
I have some of the space and time I need to be myself, but not enough.
1.4390.10014.4
• 4.
I don’t have any space or time to be myself.
0.3600.0418.8
Feeling supported and encouraged
• 1.
I feel I have the encouragement and support I want.
2.3270.15614.9
• 2.
I feel I have adequate encouragement and support.
2.2010.14515.1
• 3.
I feel I have some encouragement and support, but not enough.
1.3200.09214.3
• 4.
I feel I have no encouragement and support.
0.4740.04610.3
Domain position in the BWS task
Position 1_B0.0000.000Constant
Position 2_B−0.0980.030−3.3
Position 3_B−0.1630.031−5.2
Position 4_B−0.2240.033−6.9
Position 5_B−0.2590.035−7.5
Position 6_B−0.2830.036−7.9
Position 7_B−0.2700.036−7.4
Position 1_W0.0000.000Constant
Position 2_W0.0200.0310.7
Position 3_W0.0020.0320.1
Position 4_W0.0250.0330.7
Position 5_W0.0270.0330.8
Position 6_W−0.0160.033−0.5
Position 7_W0.0350.0331.1
Scale parameters
Scale parameter for younger participants: data relate to participants aged 18-44 years1.000Base
Scale parameter for older participants: data relate to participants aged 45+ years1.1843.1
Scale parameter for time to complete BWS task: duration ≤6 minutes (360 seconds)1.000Base
Scale parameter for time to complete BWS task: duration >6 minutes (361 seconds+)1.2503.3
Scale parameter for participants with lower educational qualifications (below BA/MA/PhD or equivalent)1.000Base
Scale parameter for participants with higher educational qualifications (BA/MA/PhD or equivalent)1.2023.4
No. of observations31 392
df42
Final log-likelihood−40 843.1
Rho2 (0)0.227
AIC81 770.1
BIC82 121.1
AIC indicates Akaike information criterion; ASCOT, Adult Social Care Outcomes Toolkit; BA, bachelor of arts; BIC, Bayesian information criterion; MA, master of arts; PhD, doctor of philosophy; SE, standard error; S-MNL, scale heterogeneity multinomial logit regression.
The attribute-level coefficients were all statistically significant compared to 'control over daily life' at level 4, similar to the findings from the original MNL model. Participants placed the highest valuation on 'occupation' at level 1 and the lowest valuation on 'control over daily life' at level 4. Position effects were also consistent with the original MNL model.
The scale parameters revealed significant variations in error variance among different groups. Findings showed those who were aged 45 years and older, those who had higher educational qualifications, and those who spent more than 6 minutes completing the BWS task made more deterministic choices and showed less error variance compared to their group counterparts.
A final model was estimated to explore variation in preferences between groups. Results for the taste heterogeneity S-MNL model are presented in Table 5.
Table 5Estimated parameters for the ASCOT-Carer measure using general population data from England taste heterogeneity S-MNL model (N = 981).
Attribute-levelTaste S-MNL
coefficientSEt-ratio (robust)
Occupation
• 1.
I’m able to spend my time as I want, doing things I value or enjoy (BA/MA/PhD or equivalent education)
2.8500.19114.9
• 1.
I’m able to spend my time as I want, doing things I value or enjoy (below BA/MA/PhD or equivalent education)
2.7840.18515.0
• 2.
I’m able do enough of the things I value or enjoy with my time
2.6200.17215.2
• 3.
I do some of the things I value or enjoy with my time, but not enough
1.4760.10314.4
• 4.
I don’t do anything I value or enjoy with my time
0.1320.0383.5
Control over daily life
• 1.
I have as much control over my daily life as I want (short-term tertiary education)
2.8280.21413.2
• 1.
I have as much control over my daily life as I want (education other than short-term tertiary education)
2.5730.17514.7
• 2.
I have adequate control over my daily life
2.2020.14814.9
• 3.
I have some control over my daily life, but not enough
1.2290.08814.0
• 4.
I have no control over my daily life
0.0000.000Constant
All levels: respondent living as married0.1740.0276.6
Looking after yourself
• 1.
I look after myself as well as I want
2.1530.14714.6
• 2.
I look after myself well enough
2.0330.14114.4
• 3.
Sometimes I can’t look after myself well enough
0.5110.05010.3
• 4.
I feel I am neglecting myself
0.2430.0416.0
Safety
• 1.
I feel as safe as I want
2.0130.13914.5
• 2.
Generally I feel adequately safe, but not as safe as I would like
1.1860.08314.2
• 3.
I feel less than adequately safe
0.6840.05612.2
• 4.
I don’t feel at all safe
0.3480.0448.0
Social participation and involvement
• 1.
I have as much social contact as I want with people I like
2.1730.14515.0
• 2.
I have adequate social contact with people
1.9490.13015.0
• 3.
I have some social contact with people, but not enough
1.2600.09213.8
• 4.
I have little social contact with people and feel socially isolated
0.4140.0468.9
All levels: respondent living as married−0.0610.026−2.3
Space and time to be yourself
• 1.
I have all the space and time I need to be myself (social grade A- high managerial, administrative or professional)
2.7420.02113.4
• 1.
I have all the space and time I need to be myself (below social grade A)
2.5430.16915.0
• 2.
I have adequate space and time to be myself
2.2910.15115.2
• 3.
I have some of the space and time I need to be myself, but not enough
1.3500.09614.0
• 4.
I don’t have any space or time to be myself
0.2600.0416.3
Feeling supported and encouraged
• 1.
I feel I have the encouragement and support I want
2.2510.15214.8
• 2.
I feel I have adequate encouragement and support
2.1230.14215.0
• 3.
I feel I have some encouragement and support, but not enough (social grade B- intermediate managerial, administrative or professional)
1.2080.09412.9
• 3.
I feel I have some encouragement and support, but not enough (all other social grades)
1.2430.09113.6
• 4.
I feel I have no encouragement and support (respondent reported as Christian (all))
0.3410.0556.6
• 4.
I feel I have no encouragement and support (everyone else: respondent reported as no religion/Buddhist/Hindu/Muslim/Sikh/Any other religion)
0.4110.0537.8
Domain position in the BWS task
Position 1_B0.0000.000Constant
Position 2_B−0.0970.030−3.2
Position 3_B−0.1620.031−5.2
Position 4_B−0.2210.033−6.7
Position 5_B−0.2550.035−7.4
Position 6_B−0.2820.036−7.9
Position 7_B−0.2680.036−7.4
Position 1_W0.0000.000Constant
Position 2_W0.0210.0310.7
Position 3_W0.0000.0320.0
Position 4_W0.0210.0330.6
Position 5_W0.0230.0330.7
Position 6_W−0.0160.033−0.5
Position 7_W0.0330.0331.0
Scale parameters
Scale parameter for younger participants: data relates to participants aged 18 years – 44 years1.0000.000Base
Scale parameter for older participants: data relates to participants aged 45+ years1.1820.0603.1
Scale parameter for time to complete BWS task: duration ≤ 6 minutes (360 seconds)1.0000.000Base
Scale parameter for time to complete BWS task: duration > 6 minutes (361 seconds+)1.2420.0733.3
Scale parameter for participants with lower educational qualifications (below BA/MA/PhD or equivalent)1.0000.000Base
Scale parameter for participants with higher educational qualifications (BA/MA/PhD or equivalent)1.2000.0583.4
No. of observations31 392
df49
Final log-likelihood−40 781.9
Rho2 (0)0.227
AIC81 661.8
BIC82 071.2
AIC indicates Akaike information criterion; ASCOT, Adult Social Care Outcomes Toolkit; BIC, Bayesian information criterion; SE, standard error.
The taste heterogeneity S-MNL model showed similar results compared to the original MNL and S-MNL models, where participants placed the highest valuation on occupation at level 1 and the lowest valuation on 'control over daily life' at level 4.
Those with higher educational qualifications placed a higher value on 'occupation' level 1 compared to those with other educational qualifications. Individuals with short-term tertiary educational qualifications also placed a higher value on 'control over daily life' level 1. The top level of the 'space and time to be yourself' attribute was also valued more highly by those who identify with social grade A (higher managerial, administrative, or professional) than those who identify with any other social grade classification. Individuals who identify with social grade B (intermediate managerial, administrative, or professional) were less concerned if they were in a situation in which they would “have some encouragement and support, but not enough” than individuals identifying with any other social grade classification. Those of Christian faith (all denominations) also placed a lower value on the bottom level of the 'support' attribute. Married individuals valued the 'control over daily life' attribute higher and the 'social participation' attribute lower compared to their unmarried counterparts.

### ASCOT-Carer Preference-Based Index Values

Population proportions were applied to certain coefficients showing evidence of taste heterogeneity and correcting for sample unrepresentativeness. The coefficients were adjusted, and the weighted average values were rescaled. (See Appendix Table 1 in Supplemental Materials found at https://doi.org/10.1016/j.jval.2019.07.014 for a comparison of rescaled coefficients [0-1 values] from the basic MNL model, the S-MNL model, and the taste heterogeneity S-MNL model.) The final preference-based index values for all attribute-levels of the ASCOT-Carer are presented in Figure 1.
We can calculate the overall SCRQoL informal carer state by summing the preference-based index values for the selected levels across each attribute. For instance, the value for state 333333 (0.405) is the sum of values at level 3 (some needs) across all attributes of the ASCOT-Carer.

## Discussion

The current study produced a set of preference-based index values for the ASCOT-Carer within a general sample from England. Respondents placed the highest value on the 'occupation' attribute at level 1 and a lower value on other attributes, with the lowest valuation on the 'control over daily life' attribute at level 4. These values can be used to assess the impact of interventions on older individuals and their informal carers in England for economic evaluations.
We found the values of the levels within each attribute of the ASCOT-Carer monotonically increased, which was in-line with our expectations given that the levels were placed on an ordinal scale. The largest utility differences were found between the second highest-valued level (level 2) compared to the second lowest-valued level (level 3) for all attributes except the 'safety' attribute. There were fewer differences between the highest and second-highest valued level (levels 1 and 2). The steep drop in perceived utility when moving between level 2 and level 3 may indicate that people tend to place higher value on positively framed outcomes (ie, ideal state or no needs), and major changes to utility are implemented once reaching a certain state (ie, some needs or high needs). Interestingly, for the 'safety' attribute, the decrements between the levels were fairly similar.
We further explored position effects on best and worst choices. Results identified that the position of the attributes in the list framed how respondents made best (first and second) choices. Respondents were more likely to indicate attributes presented at the top of the list were the best, and less likely to choose attributes as the best as they moved down the list, which is consistent with the literature.
• Campbell D.
• Erdem S.
Position bias in best-worst scaling surveys: a case study on trust in institutions.
• Netten A.
• Burge P.
• Malley J.
• et al.
Outcomes of social care for adults: developing a preference-weighted measure.

Burge P, Potoglou D, Kim CW, Hess S. How do the public value different outcomes of social care? Estimation of preference weights for ASCOT. 2010.

This framing was not apparent for worst (first and second) choices. Participants may use different heuristics and psychological processes when evaluating profiles and selecting best and worst choices.
• Mühlbacher A.C.
• Kaczynski A.
• Zweifel P.
• Johnson F.R.
Experimental measurement of preferences in health and healthcare using best-worst scaling: an overview.
For instance, respondents may be more inclined to choose the first or second positioned item for the best choices but examine each item in every position for the worst choices. Choice probability from the MNL model assumes all participants consider the process of choosing best and worst choices is the same, and this work further confirms the need to randomize the ordering of the attributes to control for the ordering effect in the BWS task.
There were also significant differences in preferences for SCRQoL in informal carers based on socioeconomic and sociodemographic characteristics, including education, marital status, social grade, and religion. Further results showed variations in error between groups. The significant age and education effects on error variance may relate to cognitive ability, which has been shown to underlie choice behavior.
• Flynn T.N.
• Louviere J.J.
• Peters T.J.
• Coast J.
Best–worst scaling: what it can do for health care research and how to do it.
• Flynn T.N.
• Huynh E.
• Peters T.J.
• et al.
Scoring the ICECAP-A capability instrument. Estimation of a UK general population tariff.
Those who took longer time to complete the BWS exercise showed less error variance compared to faster respondents. This finding is in line with previous work revealing greater error variance for quicker respondents in online stated-choice experiments.
• Campbell D.
• Mørkbak M.R.
• Olsen S.B.
Response time in online stated choice experiments: the non-triviality of identifying fast and slow respondents.
One explanation is some respondents take longer to understand the cognitive processes underpinning the BWS task, suggesting slower respondents use more cognitive effort and make more deterministic choices.
• Hawkins G.E.
• Marley A.A.J.
• Heathcote A.
• Flynn T.N.
• Louviere J.J.
• Brown S.D.
Integrating cognitive process and descriptive models of attitudes and preferences.
This includes taking into account all of the attributes in each scenario and weighing the available alternatives.
• Rose J.M.
• Black I.R.
Means matter, but variance matter too: decomposing response latency influences on variance heterogeneity in stated preference experiments.
This work further investigates the BWS exercise to assess whether it is a viable technique to value SCRQoL states in a general sample. There was good completion of the BWS exercise, and most participants understood the BWS and were able to put themselves into the hypothetical state as a carer.
• Flynn T.N.
• Huynh E.
• Peters T.J.
• et al.
Scoring the ICECAP-A capability instrument. Estimation of a UK general population tariff.
Most participants also found the exercise fairly easy to complete. The BWS exercise was administered through a web-based survey, which allowed participants to easily access and complete the questionnaire. This design also allowed us to achieve a large sample size to give sufficient power to estimate preferences and provide robust results to explore scale and taste heterogeneity.
• Huynh E.
• Coast J.
• Rose J.
• Kinghorn P.
• Flynn T.
Values for the ICECAP-Supportive Care Measure (ICECAP-SCM) for use in economic evaluation at end of life.
There are some limitations worth noting in the study. It is argued that scale and taste heterogeneity should be investigated together, but there is some debate about whether estimates from the S-MNL model are biased.

Keele L, Park DK. Difficult choices: an evaluation of heterogenous choice models. Paper presented at: 2004 Meeting of the American Political Science Association; Sept. 2-5, 2004; Chicago, IL.

The S-MNL models are routinely used to estimate preferences
• Fiebig D.G.
• Keane M.P.
• Louviere J.
• Wasi N.
The generalized multinomial logit model: accounting for scale and coefficient heterogeneity.
; however, others proposed using alternative methods to analyze the data, such as scale-adjusted latent class analysis.
• Flynn T.N.
• Huynh E.
• Peters T.J.
• et al.
Scoring the ICECAP-A capability instrument. Estimation of a UK general population tariff.
• Ratcliffe J.
• Huynh E.
• Stevens K.
• Brazier J.E.
• Sawyer M.
• Flynn T.
Nothing about us without us? A comparison of adolescent and adult health-state values for the child health utility-9D using profile case best-worst scaling.
Future work could further examine BWS data using this method to further investigate the accuracy of the method. Another limitation is some variable subgroups of the sample were unrepresentative compared to the general population (education, marital status, social grade, and religion). We aimed to account for this by adjusting the coefficients to take into account population proportions.
There are a number of implications for social care policy and practice drawn from this work. This study has filled a gap by generating preference weights for the ASCOT-Carer, an outcome measure for informal carers based on attributes that matter most to people. This expands the use of the measure, making it suitable for use in economic evaluations of interventions and support. The ASCOT-Carer was designed to capture broader well-being experiences relevant to caring for other people, rather than simply measuring health effects. This is useful for social care practice
• Netten A.
• Burge P.
• Malley J.
• et al.
Outcomes of social care for adults: developing a preference-weighted measure.
in England and could complement the preference-weighted ASCOT service user measure to understand the impact of interventions on older people and people with long-term conditions and their informal carers.

## Conclusion

We estimated a set of preference-weighted index values for the English version of the ASCOT-Carer measure from the general population in England using BWS. BWS has shown to be a viable technique to value SCRQoL informal carer states to be used for understanding the impact of social care and interventions on older people, people with long-term conditions and their informal carers in England for economic evaluations.

## Acknowledgments

This project was funded by the NORFACE Welfare State Futures program under grant number 462-14-160 . In addition, the Austrian contribution to this project was cofunded by the Austrian Science Fund (FWF, project number I 2252-G16) and the Vienna Social Fund (FSW). The Finnish contribution to this project was co-funded by the National Institute for Health and Welfare . The views expressed are not necessarily those of the funders. The authors would like to thank all participants for taking part in the current study, as well as participating organizations responsible for recruitment (Research Now). The authors also thank attendants at the ISPOR 22nd Annual International Meeting for their useful feedback and suggestions on this work.

## Supplemental Material

• Appendix Figure 1 and Appendix Table 1

## References

• Al-Janabi H.
• Van Exel J.
• Brouwer W.
• Coast J.
A framework for including family health spillovers in economic evaluation.
Med Decis Making. 2016; 36: 176-186
• Al-Janabi H.
• Van Exel J.
• Brouwer W.
• et al.
Measuring health spillovers for economic evaluation: a case study in meningitis.
Health Econ. 2016; 25: 1529-1544
• Wittenberg E.
• Prosser L.A.
Disutility of illness for caregivers and families: a systematic review of the literature.
Pharmacoeconomics. 2013; 31: 489-500
• Prosser L.A.
• Lamarand K.
• Gebremariam A.
• Wittenberg E.
Measuring family HRQoL spillover effects using direct health utility assessment.
Med Decis Making. 2015; 35: 81-93
• Pinquart M.
• Sörensen S.
Differences between caregivers and noncaregivers in psychological health and physical health: a meta-analysis.
Psychol Aging. 2003; 18: 250-267
• Pinquart M.
• Sörensen S.
Correlates of physical health of informal caregivers: a meta-analysis.
J Gerontol B Psychol Sci Soc Sci. 2007; 62: P126-P137
• Al-Janabi H.
• Flynn T.N.
• Coast J.
Estimation of a preference-based carer experience scale.
Med Decis Making. 2011; 31: 458-468
• Basu A.
• Meltzer D.
Implications of spillover effects within the family for medical cost-effectiveness analysis.
J Health Econ. 2005; 24: 751-773
• Al-Janabi H.
• Flynn T.N.
• Coast J.
QALYs and carers.
Pharmacoeconomics. 2011; 29: 1015-1023
• Goodrich K.
• Kaambwa B.
• Al-Janabi H.
The inclusion of informal care in applied economic evaluation: a review.
Value Health. 2012; 15: 975-981
• Reed C.
• Barrett A.
• Lebrec J.
• et al.
How useful is the EQ-5D in assessing the impact of caring for people with Alzheimer’s disease?.
Health Qual Life Outcomes. 2017; 15: 16
• Michels C.T.
• Boulton M.
• Wee B.
• Peters M.
Psychometric properties of carer-reported outcome measures in palliative care: a systematic review.
Palliat Med. 2016; 30: 23-44
• Harvey K.
• Catty J.
• Langman A.
• et al.
A review of instruments developed to measure outcomes for carers of people with mental health problems.
Acta Psychiatr Scand. 2008; 117: 164-176
• Shilling V.
• Matthews L.
• Jenkins V.
• Fallowfield L.
Patient-reported outcome measures for cancer caregivers: a systematic review.
Qual Life Res. 2016; 25: 1859-1876
• Deeken J.F.
• Taylor K.L.
• Mangan P.
• Yabroff K.R.
• Ingham J.M.
Care for the caregivers: a review of self-report instruments developed to measure the burden, needs, and quality of life of informal caregivers.
J Pain Symptom Manage. 2003; 26: 922-953
• Wiseman V.
Caring: the neglected health outcome? or input?.
Health Policy. 1997; 39: 43-53
• Brouwer W.B.F.
• van Exel N.J.A.
• van Gorp B.
• Redekop W.K.
The CarerQol instrument: a new instrument to measure care-related quality of life of informal caregivers for use in economic evaluations.
Qual Life Res. 2006; 15: 1005-1021
• Al-Janabi H.
• Coast J.
• Flynn T.N.
What do people value when they provide unpaid care for an older person? A meta-ethnography with interview follow-up.
Soc Sci Med. 2008; 67: 111-121
• Rand S.E.
• Malley J.N.
• Netten A.P.
• Forder J.E.
Factor structure and construct validity of the Adult Social Care Outcomes Toolkit for Carers (ASCOT-Carer).
Qual Life Res. 2015; 24: 2601-2614
• Rand S.E.
• Malley J.N.
• Netten A.P.
Measuring the Social Care Outcomes of Informal Carers: An Interim Technical Report for the Identifying the Impact of Social Care (IIASC) Study.
Canterbury, England2012 (QORU Discussion Paper 2833)
1. Smith N, Holder J, Netten A. Measuring outcomes for carers. Paper presented at: British Society of Gerontology Annual Conference; September 4-6, 2008.

• Holder J.
• Smith N.
• Netten A.
Outcomes and Quality for Social Care Services for Carers: Kent County Council Carers Survey Development Project 2007-2008.
Canterbury, England2009 (Technical report)
• Flynn T.N.
• Louviere J.J.
• Peters T.J.
• Coast J.
Best–worst scaling: what it can do for health care research and how to do it.
J Health Econ. 2007; 26: 171-189
• Flynn T.N.
Valuing citizen and patient preferences in health: recent developments in three types of best–worst scaling.
Expert Rev Pharmacoecon Outcomes Res. 2010; 10: 259-267
• Mohide E.A.
• Torrance G.W.
• Streiner D.L.
• Pringle D.M.
• Gilbert R.
Measuring the wellbeing of family caregivers using the time trade-off technique.
J Clin Epidemiol. 1988; 41: 475-482
• Whitty J.A.
• Walker R.
• Golenko X.
• Ratcliffe J.
A think aloud study comparing the validity and acceptability of discrete choice and best worst scaling methods.
PLoS One. 2014; 9: e90635
• Cheung K.L.
• Wijnen B.F.M.
• Hollin I.L.
• et al.
Using best-worst scaling to investigate preferences in health care.
Pharmacoeconomics. 2016; 34: 1195-1209
• Flynn T.N.
• Louviere J.J.
• Peters T.J.
• Coast J.
Using discrete choice experiments to understand preferences for quality of life. Variance-scale heterogeneity matters.
Soc Sci Med. 2010; 70: 1957-1965
• Flynn T.N.
• Marley A.A.J.
Best-worst scaling: theory and methods.
in: Hess S. Daly A. Handbook of Choice Modelling. Edward Elgar Publishing, 2014
• Louviere J.J.
• Flynn T.N.
• Marley A.A.J.
Best-Worst Scaling: Theory, Methods and Applications.
Cambridge University Press, Cambridge, UK2015
• Flynn T.N.
• Huynh E.
• Peters T.J.
• et al.
Scoring the ICECAP-A capability instrument. Estimation of a UK general population tariff.
Health Econ. 2015; 24: 258-269
• Ratcliffe J.
• Huynh E.
• Stevens K.
• Brazier J.E.
• Sawyer M.
• Flynn T.
Nothing about us without us? A comparison of adolescent and adult health-state values for the child health utility-9D using profile case best-worst scaling.
Health Econ. 2016; 25: 486-496
• Johnson F.R.
• Lancsar E.
• Marshall D.
• et al.
Constructing experimental designs for discrete-choice experiments: report of the ISPOR conjoint analysis experimental design good research practices task force.
Value Health. 2013; 16: 3-13
• Day B.
• Bateman I.J.
• Carson R.T.
• et al.
Ordering effects and choice set awareness in repeat-response stated preference studies.
J Environ Econ Manage. 2012; 63: 73-91
• Campbell D.
• Erdem S.
Position bias in best-worst scaling surveys: a case study on trust in institutions.
Am J Agric Econ. 2015; 97: 526-545
• Netten A.
• Burge P.
• Malley J.
• et al.
Outcomes of social care for adults: developing a preference-weighted measure.
Health Technol Assess. 2012; 16: 1-166
• Dolders M.G.T.
• Zeegers M.P.A.
• Groot W.
• Ament A.
A meta-analysis demonstrates no significant differences between patient and population preferences.
J Clin Epidemiol. 2006; 59: 653-664
• Dolan P.
Whose preferences count?.
Med Decis Mak. 1999; 19: 482-486
Conditional logit analysis of qualitative choice behavior.
in: Zarembka P. Frontiers in Econometrics. Academic Press, New York, NY1974: 105-142
• Train K.
Discrete Choice Methods in Simulation.
Cambridge University Press, Cambridge, UK2003
• Huynh E.
• Coast J.
• Rose J.
• Kinghorn P.
• Flynn T.
Values for the ICECAP-Supportive Care Measure (ICECAP-SCM) for use in economic evaluation at end of life.
Soc Sci Med. 2017; 189: 114-128
• Fiebig D.G.
• Keane M.P.
• Louviere J.
• Wasi N.
The generalized multinomial logit model: accounting for scale and coefficient heterogeneity.
Mark Sci. 2010; 29: 393-421
2. ALOGIT.
3. Bierlaire M. BIOGEME: a free package for the estimation of discrete choice models. Paper presented at: 3rd Swiss Transport Research Conference; 2003; Ascona, Switzerland.

4. Burge P, Potoglou D, Kim CW, Hess S. How do the public value different outcomes of social care? Estimation of preference weights for ASCOT. 2010.

• Coast J.
• Flynn T.N.
• Natarajan L.
• et al.
Valuing the ICECAP capability index for older people.
Soc Sci Med. 2008; 67: 874-882
• Office for National Statistics
2011 Census for England, 18+ population from England and Wales.
(Accessed August 1, 2018)
• Office for National Statistics
Analysis Tool.
(Accessed August 1, 2018)
• Office for National Statistics
Census for England, population estimates for those aged 16+ from England and Wales.
(Accessed August 1, 2018)
• Office for National Statistics
2011 Census for England, population estimated for those aged between 16 and 74 from England and Wales.
(Accessed August 1, 2018)
5. 2011 Census for England, all residents for England between 16 and 64.
(Accessed August 1, 2018)
• Office for National Statistics
Census for England, all residents from England and Wales.
(Accessed August 1, 2018)
• Mühlbacher A.C.
• Kaczynski A.
• Zweifel P.
• Johnson F.R.
Experimental measurement of preferences in health and healthcare using best-worst scaling: an overview.
Health Econ Rev. 2016; 6
• Campbell D.
• Mørkbak M.R.
• Olsen S.B.
Response time in online stated choice experiments: the non-triviality of identifying fast and slow respondents.
J Environ Econ Policy. 2017; 6: 17-35
• Hawkins G.E.
• Marley A.A.J.
• Heathcote A.
• Flynn T.N.
• Louviere J.J.
• Brown S.D.
Integrating cognitive process and descriptive models of attitudes and preferences.
Cogn Sci. 2014; 38: 701-735
• Rose J.M.
• Black I.R.
Means matter, but variance matter too: decomposing response latency influences on variance heterogeneity in stated preference experiments.
Mark Lett. 2006; 17: 295-310
6. Keele L, Park DK. Difficult choices: an evaluation of heterogenous choice models. Paper presented at: 2004 Meeting of the American Political Science Association; Sept. 2-5, 2004; Chicago, IL.