The analysis of reforms to the structure of taxes and benefits facing households has long been a popular pastime among economists. The central problem with such reforms is that they are either expensive in terms of government revenue or involve making some households better off at the expense of others. Generally, there is a trade-off between economic efficiency and equity, since the reform alters both the structure of economic incentives and the distribution of incomes across households. To evaluate this trade-off, it is necessary to quantify any incentive effects and provide an appropriate measure of the pre- and post-reform standard of living for each household. This is the role of microeconometrics.

If it can be shown that incentive effects are small, then ignoring them altogether may provide a sufficiently good approximation to reality. The reforms can then be discussed without these complications. If they are significant, however, then ignoring them will understate the gains and overstate the losses to households which the reform induces. The result could be an unduly pessimistic view of the consequences of the reform and an in-built bias favouring the status quo.

The likely size of incentive effects is therefore a fundamental issue. The question arises whether we have at hand suitable data and suitable model selection techniques for the accurate measurement of such household responses. Clearly the procedures used to develop a satisfactory microeconometric model will differ in many ways to those used in the selection of 'data-coherent' macroeconometric models. To illustrate precisely what is involved, let us turn to a specific and frequently analysed area of reform - direct taxation.

Attention in the UK and elsewhere has been focused for the most part on the reform of direct taxation rather than on the indirect taxation of expenditures or the taxation of asset income. For example, a popular proposed reform has been a negative income tax. This would involve a tax-free lump sum payment to each household, dependent on that household's characteristics, and the taxation of all earned income. These schemes have various names: social dividends, basic income guarantees or tax credits. They all share the common structure outlined above and involve the integration of the tax and income support systems. Here, as we shall claim is often the case, individuals' responses will tend to counter-balance each other. They will have only a negligible effect in aggregate on government revenue but a significant effect on the distribution of income and employment.

Such schemes have actually been implemented in the United States but only as temporary experiments in particular localities designed to measure the likely size of individual responses to a negative income tax. In the UK, a negative income tax scheme was proposed in the 1972 Green Paper, 'Proposals for a Tax Credit Scheme', but its central proposals were never implemented.

More recently, there have been proposals to alter the income taxation of couples. In the UK, the married man's tax allowance has been strongly criticized, following the publication of the 1980 Green Paper on 'The Taxation of Husband and Wife'. Here the question is one of the appropriate distribution of the tax burden between couples with and without children, and between couples with and without a working wife. The structure of the UK tax system has again remained unchanged despite arguments for reform.

To analyse the effects of these reforms, we need to know how they affect the labour market behaviour of households. Since the taxation of husband and wife seems to be a central issue, our empirical work has focused on family labour market behaviour using the large household surveys available in the UK. The results of this analysis, although illustrative, show that such reforms sometimes have dramatic effects on labour market behaviour. Furthermore, where responses do occur, they tend to be concentrated in the labour market decisions of women. Male labour market behaviour seems to display only small predictable responses. Different reforms clearly have quite different effects however, especially with regard to the participation decision and the choice between full- and part-time work.

Naturally, any microeconometric model worth its salt should assess the importance of macroeconomic variables on the supply of job offers and the ability of women to adjust their hours of work. Models are necessarily estimated from past observed behaviour under past labour market conditions. In order to predict current behaviour under current conditions, we need to incorporate these macroeconomic effects explicitly in our model. The most satisfactory way we have found to do this is to scale the participation probability by a term reflecting the probability of a job offer being made to a particular individual in a given location with specific skills. This remains, however, a crucial area for further research.

Models estimated from individual or household data raise additional problems. In contrast to macroeconomic modelling, it is very important to choose the correct form for the distribution of the unobservable variables which reflect the wide variety of indivual attitudes to the nature of the paid work available. Consequently, we need a rather different set of model selection 'diagnostics' - tests which check the shape of this distribution and its relationship to observable factors. Panel data - data from surveys that follow the same individuals through time - would be even more valuable in identifying and eliminating these unobservable individual effects, thereby improving the robustness of our results. But no such panel data exist in the UK and this presents a severe drawback for the analysis of labour market responses. It is difficult to overstate the advantages that panel data would offer for the type of empirical microeconomics described here.

Finally, it is worth noting that simple models which impose constant elasticities, constant income effects or restrictive substitution effects are rarely satisfactory. They often unknowingly eliminate certain types of quite plausible behaviour (for example, the joint occurrence of forward and backward bending supply curves) and nearly always over-emphasize other types of behaviour. Moreover, most are firmly rejected by the data and therefore do not adequately reflect the wide scope of behaviour observed in household surveys. The reason for choosing simple models is often the ease with which they can be used for policy simulation. This is now an unnecessary limitation, however, since we and other researchers have recently generated efficient methods for simulating complex reforms using models that have none of these restrictive properties.

As an illustration, let us consider a particular reform - the abolition of the married man's allowance and the payment of a higher child benefit for younger children to women who are not in the labour market. Suppose that the amount of this child benefit is increased so that in the absence of any household responses, the reform would leave government revenue unchanged. The potential impact on behaviour is quite revealing. There is a clear incentive for mothers who work relatively few hours to leave the labour market altogether, and for those remaining (in particular those with working husbands) to increase their working hours. This is indeed what our model predicts, and it generates a significant impact on the distribution of income and employment. The two responses have off-setting effects on government revenue, however, and the reform remains approximately neutral in aggregate terms even after allowing for labour market responses. This underlines the obvious importance of analysing behaviour at the micro level rather than using aggregate data.

In conclusion, there is a clear need for further research and collaboration between research groups in this area, especially for the development of a full range of model selection criteria with properties that are well understood. In general, the links between economic theory, econometric practice and policy evaluation make this an exciting area for new research.

Richard Blundell

This is one of a series of articles describing research relevant to economic policy undertaken by CEPR Research Fellows. Richard Blundell is Professor of Political Economy at University College London and an Associate Director of the Developments in Applied Economic Theory and Econometrics programme at CEPR. The research referred to in this article is contained in two papers written jointly with Costas Meghir (University College London), Liz Symons (Manchester University) and Ian Walker (Manchester University) with the active encouragement of colleagues at the Institute for Fiscal Studies. The first of these two papers deals with more technical issues and is available as UCL discussion paper No. 84-13. The second paper describes simulations of the effects of various tax reforms and is forthcoming in the January issue of Fiscal Studies. Further information concerning the research discussed in this article may be obtained by contacting Richard Blundell at CEPR.