Environmental Audit CommitteeWritten evidence submitted by Professor Lord Richard Layard, Centre for Economic Performance (Wellbeing Programme), LSE

How can we base public policy on subjective wellbeing?

There is a widespread desire to measure subjective wellbeing: “if you treasure it, measure it”. But how shall we use the information? And what other information would we need to make subjective wellbeing (SWB) an effective policy goal?

To answer these questions we need:

(i)a new type of cost-benefit analysis, and

(ii)a clear concept of how wellbeing is determined.

Let me discuss both issues, illustrating the kinds of new policy priorities which might emerge.

1. The New Cost-Benefit Framework For Policy

Public policy is of course made in many ways, some more formal than others. But it always helps to start with the ideal case, even if it cannot be used all the time. In current practice the ideal method of policy-making is by cost-benefit analysis where the units of measurement are dollars. For each person affected by a policy change we attempt to evaluate his change in wellbeing in terms of how much he would be willing to pay to bring about the change. In other words, changes are evaluated in terms of their impact on the present value of GDP, properly measured. This works quite well where we can infer willingness-to-pay from choice behaviour. So it is of obvious value in relation to education, transport policy, industrial policy and some aspects of the environment.

But the majority of public expenditure goes on quite different areas: health, social care, law and order, employment and of course the redistribution of income. (I omit defence which is especially difficult to analyse under any framework). In these other areas politicians have always implicitly used judgements about the wellbeing effects of these policies.

In health, Britain and some other countries already try to evaluate policy in terms of its impact on Quality Adjusted Life Years (QALYs). This measure is not very satisfactory since the quality-of-life impact of a disease is obtained by asking the general population how many years of life they would be willing to give up to avoid the disease. Such hypothetical questions have been shown to give very poor estimates in many contexts.1

Now however the new science of wellbeing enables us in principle to obtain direct measures of the impact of a policy change—measured in units of subjective wellbeing.

So thinking about policy change should proceed in 4 steps:

(1)First we look at the distribution of wellbeing in a country or a city, observe where wellbeing is low or where it is obviously lower than it should be;

(2)We try to understand the process which has brought about these outcomes (see next section);

(3)We invent possible policies which might deal with the worst outcomes;

(4)We evaluate these policies as a guide to which to adopt.

The evaluation has two elements: benefit and cost. In principle each of these can accrue to individuals, businesses or taxpayers. In practice, it is useful to assemble on the one side the net benefits to the private sector and on the other side the net costs to the government. In the new model the net benefits would always be measured in units of SWB.2

The net costs to the government would need careful modelling to include all the indirect cost savings (or the opposite) which the gross cost might generate. These taxpayer costs would initially be measured in dollars.

Thus benefits and costs would be in different units. But this does not matter if public expenditure is pre-determined. In this case the issue is for all possible projects i to

(1)

Assuming no problems with indivisibilities, this leads to the following rule:

Select all projects i for which

(2)

This is cost-effectiveness analysis. But it does not of course help us to reflect on the level of total public expenditure. For this we have to estimate the effect on SWB of the taxes used to finance the public expenditure cost. This can in principle be estimated from an SWB

Figure 1

equation in which disposable income appears as an independent variable, thus yielding an estimate of the marginal utility of income. This coefficient will vary with income, so let us call it λ (Y). Then, ignoring income distribution, we should select projects in order of their Bi/Ci until for the marginal project j.

(3)

This position is illustrated in Figure 1. However it is unlikely that governments will be willing to let public expenditure be determined in this way.

Moreover, even if we stick with cost-effectiveness analysis, there are still three substantial problems:

(i)the distribution of wellbeing

(ii)the discount rate, and

(iii)life expectancy.

Distribution of Wellbeing

For Jeremy Bentham there was no problem about the distribution of wellbeing: we should simply maximise its average level. But modern ethical analysis rejects this point of view. For example, if we imagine individuals judging different states of the world, not knowing which participant they will be, they would almost certainly show some degree of risk-aversion in relation to wellbeing itself.3 However citizens (and politicians) differ in their degree of concern for inequalities in wellbeing. The best that analysts can do is to illustrate the implications of different degrees of inequality aversion.

Discount Rates

When it comes to discount rates, things are more clear-cut. There can be no ethical reason for supposing that the SWB of a future generation is less important than our own. Of course, if their SWB is higher than ours, then we may attach a lower marginal value to extra units of SWB for them, as compared with us. And, climate sceptics please note, vice versa. But this is nothing to do with the present versus the future.

The preceding remarks relate to the aggregation of benefits. On the public expenditure side, if this is constrained separately in each period, there will of course be in principle a separate shadow price for expenditure in each period.

Length of Life and Number of Lives

As regards the number of people, I would argue that there is no intrinsic merit in having more or fewer lives. It is the quality of the lives which actually occur which matters. So too does their length. The standard practice is to multiply the average quality of a person’s life by the number of years it is lived. There is no obviously better alternative.

So we come now to the issue of evidence: how can we know what benefits a policy will bring?

2. The Evidence Base

As I have said, there are three main stages in policy formulation:

(i)Noticing and understanding the problem.

(ii)Design of possible solutions.

(iii)Evaluation of each solution.

The measurement of population wellbeing is certainly the first step in (1). These types of cross-sectional data have already shown us the importance for SWB of employment status, job characteristics, income, family status, health, local environment, age and gender. These analyses are extremely interesting and have highlighted the policy importance of full employment, of well-designed jobs, of family life, of adequate income, and of health, especially mental health.

But they leave many questions unanswered. This is obvious when we ask ourselves: What would we do to make things better for these people? How many of these problems would be best affected by directly helping the person now, and how many would in fact be handled best by earlier intervention in people’s lives? In many cases it would be more effective (and less demeaning) to intervene earlier—which is of course why so much public expenditure is focussed on childhood.

Wellbeing over the Life-course

So for policy purposes we need a model of SWB over the life-course, which includes a person’s present circumstances but also an explanation of how they got to that position. Let me give an example based on the British Cohort Study of people born in 1970.

The individual is born into a family with a certain socio-economic level (income, education, SEG) and certain psychological characteristics (maternal health, family stability). She then develops an emotional health of her own, a pattern of pro/anti-social behaviour and a level of intellectual development. These are measured at ages five, 10 and 16, at each of which points we could envisage policy intervention. Then by the age of 34 the person is an adult with a given level of emotional health, criminal record, educational qualifications, employment, earnings, family structure and physical health. What happens at each stage is in principle affected by everything which goes before—it is a recursive, path model.

The outcome of interest is life-satisfaction at 34. There are three ways of explaining this:

Method 1 is similar to what will emerge from the OECD’s recommended surveys. We explain life-satisfaction by the other current features of life at age 34.

Method 2 is the crude life-course approach when life-satisfaction is explained directly by the person’s childhood and family background.

Method 3 is the most illuminating where life-satisfaction is explained by the other features of life at 34 (as in Method 1) but then each of these is explained by childhood and family background. This is much the most informative. It gives almost exactly the same role to childhood as Method 2 but explains more clearly how childhood affects our life-satisfaction as adults (see Annex).

Figure 2

A MODEL OF THE LIFE-COURSE

Policy Analysis and Experimentation

So how does this help in policy analysis? It certainly suggests areas where we may want to intervene. But the only really convincing way to evaluate a policy change is through experimentation with a proper control group. The problem with experiments is that if the experiment is at age 10 and we want to know its effects at age 34 we have to wait 24 years. Models can help us overcome this problem. If we know the effect size of an intervention at age 10, we can use the model to simulate its effects at age 34.

Of course this will only work if the model is truly causal. The main problem with most cohort models (and most social science) is that they omit the genes. If emotional wellbeing at 10 appears to affect emotional wellbeing at 34 with a partial correlation of .10, this is partly because both variables are correlated with a common genetic influence. So if we altered wellbeing at 10 it might well affect wellbeing at 34 by less than appears from most models of the life-course.

Ultimately we must use cohort models with either DNA or twin data. But in the meantime we can use those we have, provided we do it with proper caution. And we shall certainly find that the cross-sectional data which the present OECD exercise throws up will be far more useful if they are combined with data from cohort studies.

Annex4

A MODEL OF MALE WELLBEING OVER THE LIFE-COURSE

(PARTIAL CORRELATION COEFFICIENTS)

Model 1. Explanation by Adult Variables

Life-satisfaction at 34:

= .17 Emotional health at 26

+ .06 Criminal record 16–34

+ .02 Educational level

+ .22 In full-time work at 34

+ .12 Earnings (if in full-time work) at 34

+ .10 Married or cohabiting at 34

+ .06 Self-perceived health at 26

Model 2. Explanation by Childhood and Background

Life-satisfaction at 34:

= .09 Family background (economics)

+ .03 Family background (psychological)

+ .25 Emotional health at 5, 10 and 16

+ .07 Good conduct at 5, 10 and 16

+ .06 Cognitive performance at 5 and 10

Model 3. Combined Model

First we estimate an equation to explain life-satisfaction at 34 which includes all variables for Model 1 and Model 2. The effects of the current variables are almost as in Model 1 but the effects of the childhood variables are negligible.

Thus childhood effects are therefore working mainly via the current variables. The following table shows these effects.

Next we confirm that combining this table with the coefficients in Model 1 provides a predictive equation similar to that in Model 2.

However the combined model is to be preferred because it explains how Model 2 occurs, giving better clues on all the effects (especially on costs) associated with any intervention.

STRUCTURAL EQUATIONS FOR ADULT SUCCESS

(PARTIAL CORRELATION COEFFICIENTS)

	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
VARIABLES	Emotional health	Good conduct	Education	Earnings (FT men only)	FT work (men only)	Having a family	Health (self-perceived)	Life-satisfaction
Family Economic	0.079***	0.033***	0.318***	0.253***	0.027***	0.024***	0.094***	0.086***
	(0.011)	(0.005)	(0.015)	(0.022)	(0.006)	(0.007)	(0.018)	(0.019)
Family Psycho-	0.093***	0.056***	0.052***	-0.025	0.015	0.033**	0.071***	0.049***
	(0.020)	(0.016)	(0.015)	(0.018)	(0.016)	(0.015)	(0.024)	(0.020)
Emotional health (5 10 16)	0.380***	-0.066***	0.014***	0.098***	0.047***	-0.027	0.193***	0.208***
	(0.023)	(0.013)	(0.003)	(0.028)	(0.016)	(0.017)	(0.023)	(0.024)
Good conduct (5 10 16)	0.107***	0.232***	0.147***	0.040***	0.059***	0.120***	0.069***	0.105***
	(0.020)	(0.023)	(0.014)	(0.010)	(0.016)	(0.024)	(0.018)	(0.020)
Cognitive Perf (5 10)	0.084***	0.060***	0.329***	0.176***	0.055***	0.084***	0.059***	0.047***
	(0.015)	(0.012)	(0.013)	(0.017)	(0.013)	(0.017)	(0.015)	(0.012)
Female	-0.302***	0.424***	-0.001			-0.041*	-0.084***	0.091***
	(0.021)	(0.018)	(0.018)			(0.024)	(0.022)	(0.022)
Observations	8,254	10,918	10,575	3,007	4,242	6,896	8,260	8,868
Adjusted R-squared	0.271	0.319	0.270	0.097	0.008	0.020	0.064	0.087

Robust standard errors in parentheses

*** p<0.01, ** p<0.05, * p<0.10

17 June 2013