Here are the comments that I sent as a response to Morocco. Unfortunately footnotes don't transfer, so this is just the text. Thank you very much for your letters and interest in OPHI’s work and particularly in the new Multidimensional Poverty Index (MPI) that we constructed with the United Nations Development Program’s Human Development Report Office for the 2010 Human Development Report to be launched as an experimental series on 4 November 2010. The MPI, like any other national poverty measure, is shaped both by constraints of data and technology, and by the purposes of the exercise. The purpose of this exercise was to come up with a measure of multidimensional poverty that could complement income poverty measures with direct measures of deprivation, and provide, like income poverty, a snapshot of acute multidimensional poverty for different data sets. Our work benefitted greatly from discussions with the HDRO team, with statistical and expert advisors to the UNDP HDRO, as well as with other colleagues including participants in an OPHI workshop in June this year. The research and applications related to multidimensional poverty measurement are developing a new degree of transparency and rigour, and we hope that this exchange can contribute to that. As the series is being introduced on an experimental basis, we do plan to improve and revise as appropriate in response to comments and feedback. You have sent several sets of comments, which are helpful and I attach a response to each of the points raised. 1. Data Constraints. First, you note that the MPI is limited by data constraints. This is a point stressed repeatedly in my paper with Maria Emma Santos, where we write, “The...binding constraint is whether the data exist. Due to data constraints (as well as, perhaps, interpretability) we have had to severely limit the dimensions. For example, we do not have sufficient data on work or on empowerment” (p 12, section on ‘Choosing Dimensions’). Data constraints affect not only the dimensions we include – as you mention – but also our ability to measure the existing dimensions (health, education, and standard of living) fully. For example the DHS and MICS surveys are not designed to measure health for all household members, but certain key variables for women of reproductive age, and children (which they do well). Our aim was to make the best possible use of existing data, while also drawing attention to data limitations (as many others have also) and undertaking robustness checks. OPHI are also committed to measuring non-traditional dimensions of poverty; in fact our other theme of research, called Missing Dimensions, seeks to develop brief survey modules on dimensions such as informal work, safety from violence, empowerment, and social connectedness, because these are often cited as features of poverty by participatory work with poor people, and clearly analysis of these features could enrich the multidimensional poverty literature. Which of these missing dimensions should be incorporated in a multidimensional poverty measure depends upon the purpose of the index as well as the associations among indicators. 2. Dynamic Context. You argue that ‘The variables on which the measurement of the MPI is based become problematic when poverty is put in a dynamic approach’ and give the example of a household who becomes non-poor because an out-of-school child passes the age of 14, so is no longer considered ‘out of school’ because he or she is too old. This may be a problem if the MPI were used with panel data. However, the DHS, MICS, and WHS surveys that we use are not panel data surveys: that is, they do not follow the same household across different periods in the way you mention, nor do they develop chronic poverty measures to track entries into and out of multidimensional poverty in different periods. Rather, they take a representative sample of the population at each point in time, and look at the relevant population (e.g. school-aged children) at that point in time. Clearly if there were a stark demographic shift (a drop in fertility meaning that in one sample the percentage of children was much lower than in the previous period), this would register in the MPI, but the kinds of shift you mention are not captured by our data sources. The forthcoming paper by Apablaza, Ocampo and Yalonetzky, which is drawn on in Alkire & Santos section 4.6, studies changes in the MPI for 10 countries across time. As we show in our paper using the cases of Ethiopia, Bangladesh and Ghana, dynamic time series studies of MPI over time can be used to analyse to different features of multidimensional poverty reduction – such as the extent to which intensity of poverty declines as well as the headcount, and which indicators drive poverty reduction (or worsen even though poverty overall goes down). The decomposition of changes allows us to assess the relative importance of changes in the headcount and the intensity in "explaining" the overall change in the adjusted headcount. Also, we can decompose the changes in the headcount to ascertain the main drivers of that change, e.g. whether it was due more to changes in the population composition (e.g. across regions), or to changes in the specific headcounts of the groups into which the headcount has been decomposed. Similarly, in the case of the average deprivations of the poor, i.e. the intensity, it is possible to express the change as a sum of changes in the proportion of the poor who are deprived in each specific dimension or variable. 3. Threshold. You argue that ‘the measurement of poverty according to the Multidimensional Poverty Index is based on a subjective threshold and does not take into consideration, therefore, comparisons with the monetary approach, the threshold of which is determined objectively.’ You also refer to MPI cutoff as arbitrary in constract to income poverty. It seems that there is some misunderstanding here. The MPI poverty cutoff – which requires people to be deprived in 30% of the dimensions in order to be identified as multidimensionally poor – is not subjective. Subjective poverty lines are set when people are asked how much they require in order not to be poor. Those working on poverty measurement use the word ‘arbitrary’ in a specialised way which is not its normal usage. A poverty line (both in the income and in multidimensional space) is referred to as ‘arbitrary’ not because it has no normative or objective basis, but rather because it is not theoretically determined. Thus an income poverty line based on a calorie basket is also often referred to as ‘arbitrary’, since it usually uses particular set of equivalence scales, assumes that market prices are right and that each person in each location can convert income into the minimum basket of goods and services required to lead a non-impoverished life. The standard practice for (arbitrary) income poverty lines is to test how sensitive are the results to the choice of the line, using stochastic dominance tests. Similarly, what we did was to compare the poverty estimate between all possible pairs of countries using a poverty cutoff of 20%, 30% and of 40%. This is detailed in Section 4.8 of the paper. We found that in 95% of the pair wise comparisons, one country has higher (lower) poverty than the other regardless of the poverty cutoff. These results suggest that the particular cutoff of 30% we use for the MPI is not a critical choice that dramatically affects results. Finally, you compare Morocco to other countries which have higher $2/day income poverty rates and higher inequality. In our view, this is precisely the information that is of interest. The MPI cannot include income because of the fact that income data are not collected in the surveys we use. In these circumstances, it will be useful to study in greater depth than has been done to date countries’ different performances on income poverty and MPI. 4. Time Periods. You write, ‘Data reference periods range from 2000 to 2008, making the classification of countries according to the Multidimensional Poverty Index groundless.’ In fact, we do not classify countries at all. You are absolutely right that the base years differ for the MPI. As we wrote, “although we would have liked to estimate poverty for exactly the same year in all countries to enable a strict cross-country comparison, this was not possible given that the different surveys have been performed in different years in each country.” (page 20). In our countries, 52% of the 5.2 billion people we cover, living in 65 countries have data that is 2005 or later; data for 44% of people living in 31 countries come from 2003 or 2004 (largely due to the World Health Survey 2003 data, which includes China), and data for 3% of people living in 10 countries come from 2000-2002. Does this make all comparisons groundless? Actually, the use of data from different years is unfortunately common, because poverty surveys are not implemented every year. Base years also differ for the HDI education indicators, the HPI indicators, the MDGs, and income poverty indicators. The methods used to deal with this vary. The 2004 Chen and Ravallion statistics use the closest survey to the reference year (which in their case is a narrower period than ours), and interpolate progress (Chen and Ravallion 2008). We considered this option. Methods have been used to predict changes in individual MDG indicators, including those related to variables we use. However such methods are also subject to criticism and debate, because they are based on a set of assumptions that may be controversial. Thus we decided to use actual data only in this 2010 MPI: not to interpolate between years, nor to use proxies for missing variables (on which please see below). The advantage of doing so is transparency: the years and indicators used can be seen and verified by anyone, and the MPI can be easily re-constructed (as indeed HCP have done) because the datasets as well as our methodologies are publicly available. The other concern that you raise in this section appears to be misplaced. You argue that in the MPI, “missing indicators are replaced by the lower or higher bounds of variables or their proxies”. This is not accurate. We did not impute any data, nor use any proxies. If a country survey did not include a variable, then the remaining variables for that dimension were used, and the weights were adjusted such that each dimension continued to receive equal weights. As in the case above, we could have chosen a different approach – the techniques exist and are feasible – but, in consultation with the UNDP’s statistical advisors, chose a more direct route instead. 5. More up-to-date Data. In a separate letter, you drew attention to your comparison of the MPI using a 2007 dataset for Morocco and argued that “it is unfair and unjust to release in 2010 reuslts based on data going back to 2003/2004.” I wrote back immediately requesting access to the dataset mentioned, but have not received a reply. We have also checked thoroughly online but no more up to date dataset is available for Morocco. OPHI are only able to construct multidimensional poverty indices using publicly available data. For example, we are aware that Uganda and Ethiopia also have more up to date DHS surveys, however these were not available to the public when we calculated the MPI hence we used the previous survey data that were available. When new data become publically available we will calculate the updated MPIs. 6. National vs International Measures. Also in your letter to me, you argue that ‘the choice of dimensions and indicators must be based on the preferences and priorities of the population. But this of course depends on the conditions and specificities of the population. This makes international comparisons irrelevant and misleading.’ There is a long and distinguished debate about the appropriateness of internationally – or for that matter nationally – comparable poverty measures, versus more context-specific, participatory measures. Clearly different kinds of measures can be of tremendous value in different contexts, and no one is suited to all purposes. The MPI is an international index, which is seeking to compare very basic aspects of acute multidimensional poverty – malnutrition, child mortality, households with no member having 5 years of education, children out of school, not having electricity, clean drinking water, sanitation, a floor, clean cooking fuel, and not owning more than one of the following assets: television, radio, telephone, bicycle, motorcycle, or refrigerator. Despite its evident limitations, it is of interest that while in some countries over half of the population are deprived in 30% of the weighted indicators, in other countries many fewer people experience this kind of acute poverty and indeed in Slovenia and Slovakia we did not identify any persons as experiencing acute multidimensional poverty. This gives a strong policy message of hope: such acute deprivation can be eradicated. Also while in many countries, like Morocco, the MPI headcount is higher, there are also cases showing that the converse is possible. Indeed this is the case for about one-fourth of the countries including Tanzania and Uzbekistan. Of course our hope and expectation is that in the coming years the MPI will improve as more data are collected. Alongside this, national governments are developing and will develop national measures which are both more accurate to their context, and which provide appropriate information for an adequate policy response. Morocco has been a leader in this work, with the innovative work of the HCP; Mexico is another country in which the official national poverty measure is now multidimensional. Hence just as alongside the $1.25/day measures countries generate national poverty lines and national poverty measures which are reported in national and international reports, it is possible that a similar evolution will occur with multidimensional poverty measures. Part of our hope is to support and learn from national and subnational initiatives as countries develop robust and transparent measures of multidimensional poverty that are tailored to their own context and that endure over time.