Good quality aid statistics are essential in order understand how, when, where and for which reasons aid is provided, and in order to properly evaluate the effects of aid. How reliable are the data currently available? In a new EBA working paper, “Reclassification or Reprioritisation? The Sector Allocation of Swedish Official Development Assistance 1973-2013”, Ulrika Ahrsjö attempts to assess the time consistency of the sectoral division of Swedish bilateral official development assistance (ODA).
Imagine an aid-supported micro credit organisation targeting women in a rural village. The organisation offers loans (mainly for small-scale services and agriculture), a savings facility, and a series of various “how to”-seminars (to write a business plan, elementary book-keeping, piggery, mushroom-growing etc.). Forced to assign one major purpose with this project, should one choose to define it as support to a semi-formal financial intermediary (24040), agricultural development (31120), vocational training (11330), or as support for an institution or organisation working for women’s empowerment (15170)? Would the project be more likely to be assigned one of these codes if the government had declared as a political priority to focus on business creation, on education, or on gender equality?
The numbers within brackets above are all purpose codes in the OECD/DAC Creditor Reporting System (CRS) data base (with the first three digits denoting the supported sector). This data is the main source of internationally comparable statistics of Official Development Assistance (ODA).
Thus, a good understanding of ODA is heavily dependent on the aggregation of individual administrators’ coding of individual projects. That this coding is non-trivial is evidenced by the guiding examples at the OECD website (accessed 2016-11-25), for example, “Privatisation of the National Energy Agency: The appropriate code is “energy policy and administrative management” (23010) and not “privatisation”. State enterprise restructuring programme: The appropriate code is “privatisation” (25020)”.
The central question in the working paper is whether the coding of projects is susceptible to political (changes in) priorities, or, in other words, whether re-formulated political goals for development assistance lead to reclassifications of individual aid projects, rather than actual reallocations of funds.
Addressing this question is presumably facilitated by the work of development innovation lab AidData that in recent years has undertaken the mammoth project of reassessing the coding of projects in the CRS database. The resulting data set is in theory more consistent than the original data, since the AidData staff are less likely to be affected by political pressure from donor country governments, and since the coding is done by a limited group of people over a limited time period. If one accepts these premises for analysis, differences in a subsequent comparison between the original sector allocation of bilateral aid and the recoded data could be attributed to inconsistencies in the CRS data, and such differences following political reformulations of ODA goals could be attributed to “political responsiveness” in coding.
Overall comparisons using the (three-digit) sector codes, as well as the (five-digit) purpose codes show that substantial differences do exist. Out of the 55,330 Swedish aid projects recorded in the CRS database, AidData has managed to reassess 29,640 projects. Out of their total monetary amount, 17 % of the funds (27 % of the projects) have received a different three-digit sector code by AidData, than the original one set by Sida. At the five-digit level the corresponding share in terms of money is 39 %. However, these represent over 50 % of the total number of projects, indicating that projects with smaller committed funds are more often being recoded.
As it turns out, methodological issues form a major constraint to the working hypothesis. In particular, while it is not clear whether AidData has access to more information than what’s in the CRS data set (where the full description can be as limited as “Events, Swe Dance”), over 20 % of the recoded cases at the sector level are assigned code 998, Unallocated/Unspecified. If this is due to insufficient documentation, the contextual knowledge of the original coders may imply a higher quality and precision of the original data and would thus invalidate any conclusion that differences are due to political responsiveness in coding.
In addition, the sheer number of codes in the CRS system may be a source of confusion, for coders of both data sets alike. For example, the main reason for differences within the health sector is that projects coded as “Basic Health” in the CRS data are coded as “Health, General” by AidData, and vice versa. While the many different sectors and sub-sectors are meant to enrich the data, it may also impair the quality of data in some instances.
As for the main question, some tentative support for a reclassification, rather than actual reallocation, of funds is found in the purpose code Democratic participation and civil society where CRS funds change differently from AidData concurrently with the increased political focus on human rights.
Another interesting finding is Sida’s use of the so-called policy markers for gender equality and environmental projects. The marker for ”significant objective” (1 on a scale from 0 to 2) for gender issues is assigned to half of all projects in later years, and for environmental questions assigned to about one in four projects. Whether or not this is motivated by the usage of the funds is best left unsaid here, but could warrant further study.
Finally, the overall lack of evidence for political responsiveness in coding should not be interpreted as proof that Swedish bilateral aid data is consistent over time. We understand the assignment of purpose codes to be a non-trivial task, probably influenced by contextual understanding as well as individual preferences. That political priorities would sometimes be pivotal in this process is not unlikely. Ideas on how to further study this question are welcome.
Ulrika Ahrsjö and Jan Pettersson