How much de­vel­op­ment data is enough?

Financial Mirror (Cyprus) - - FRONT PAGE -

Rapid ad­vances in tech­nol­ogy have dra­mat­i­cally low­ered the cost of gath­er­ing data. Sen­sors in space, the sky, the lab, and the field, along with new­found op­por­tu­ni­ties for crowd­sourc­ing and wide­spread adop­tion of the In­ter­net and mo­bile tele­phones, are making large amounts of in­for­ma­tion avail­able to those for whom it was pre­vi­ously out of reach. A small-scale farmer in ru­ral Africa, for ex­am­ple, can now ac­cess weather fore­casts and mar­ket prices at the tap of a screen. This data revo­lu­tion of­fers enor­mous po­ten­tial for im­prov­ing de­ci­sion-making at ev­ery level – from the lo­cal farmer to worldspan­ning de­vel­op­ment or­gan­i­sa­tions. But gath­er­ing data is not enough. The in­for­ma­tion must also be man­aged and eval­u­ated – and do­ing this prop­erly can be far more com­pli­cated and ex­pen­sive than the ef­fort to col­lect it. If the de­ci­sions to be im­proved are not first prop­erly iden­ti­fied and an­a­lysed, there is a high risk that much of the col­lec­tion ef­fort could be wasted or mis­di­rected.

This con­clu­sion is it­self based on em­pir­i­cal anal­y­sis. The ev­i­dence is weak, for ex­am­ple, that mon­i­tor­ing ini­tia­tives in agri­cul­ture or en­vi­ron­men­tal man­age­ment have had a pos­i­tive im­pact. Quan­ti­ta­tive anal­y­sis of de­ci­sions across many do­mains, in­clud­ing en­vi­ron­men­tal pol­icy, busi­ness in­vest­ments, and cy­ber se­cu­rity, has shown that peo­ple tend to over­es­ti­mate the amount of data needed to make a good de­ci­sion or mis­un­der­stand what type of data are needed.

Fur­ther­more, grave er­rors can oc­cur when large data sets are mined us­ing ma­chine al­go­rithms with­out hav­ing first hav­ing prop­erly ex­am­ined the de­ci­sion that needs to be made. There are many ex­am­ples of cases in which data min­ing has led to the wrong con­clu­sion – in­clud­ing in med­i­cal di­ag­noses or le­gal cases – be­cause ex­perts in the field were not con­sulted and crit­i­cal in­for­ma­tion was left out of the anal­y­sis.

De­ci­sion science, which com­bines un­der­stand­ing of be­hav­iour with univer­sal prin­ci­ples of co­her­ent de­ci­sion-making, lim­its th­ese risks by pair­ing em­pir­i­cal data with ex­pert knowl­edge. If the data revo­lu­tion is to be har­nessed in the ser­vice of sus­tain­able de­vel­op­ment, the best prac­tices of this field must be in­cor­po­rated into the ef­fort.

The first step is to iden­tify and frame fre­quently re­cur­ring de­ci­sions. In the field of de­vel­op­ment, th­ese in­clude large-scale de­ci­sions such as spend­ing pri­or­i­ties – and thus bud­get al­lo­ca­tions – by gov­ern­ments and in­ter­na­tional or­gan­i­sa­tions. But it also in­cludes choices made on a much smaller scale: farm­ers pon­der­ing which crops to plant, how much fer­tiliser to ap­ply, and when and where to sell their pro­duce.

The sec­ond step is to build a quan­ti­ta­tive model of the un­cer­tain­ties in such de­ci­sions, in­clud­ing the var­i­ous trig­gers, con­se­quences, con­trols, and mit­i­gants, as well as the dif­fer­ent costs, ben­e­fits, and risks in­volved. In­cor­po­rat­ing – rather than ig­nor­ing – dif­fi­cult-to-mea­sure, highly un­cer­tain fac­tors leads to the best de­ci­sions.

When put in the ser­vice of sus­tain­able de­vel­op­ment, such a model will of­ten in­volve pro­ject­ing the im­pact of in­ter­ven­tions on liveli­hoods and the en­vi­ron­ment over sev­eral decades. This process is suc­cess­ful when stake­hold­ers and ex­perts iden­tify the rel­e­vant vari­ables and their re­la­tion­ships. Par­tic­i­pants must be trained to pro­vide quan­ti­ta­tive es­ti­mates of their un­cer­tainty for the dif­fer­ent vari­ables. For ex­am­ple, ex­perts might es­ti­mate with 90% con­fi­dence, based on avail­able data and their own ex­pe­ri­ence, that farm­ers’ av­er­age maize yields in a given re­gion are 0.5-2 tons per hectare.

The third step is to com­pute the value of ob­tain­ing ad­di­tional in­for­ma­tion – some­thing that is pos­si­ble only if the un­cer­tain­ties in all of the vari­ables have been quan­ti­fied. The value of in­for­ma­tion is the amount a ra­tio­nal de­ci­sion-maker would be will­ing to pay for it. So we need to know where ad­di­tional data will have value for im­prov­ing a de­ci­sion and how much we should spend to get it. In some cases, no fur­ther in­for­ma­tion may be needed to make a sound de­ci­sion; in oth­ers, ac­quir­ing fur­ther data could be worth mil­lions of dol­lars.

This process is re­peated un­til there is no fur­ther value in ac­quir­ing data and a sound de­ci­sion – a log­i­cal con­clu­sion, based on the in­for­ma­tion, val­ues, and pref­er­ences of the de­ci­sion-makers or de­ci­sion-making body – is reached. It pro­vides de­ci­sion-makers and stake­hold­ers in­sights into how to im­prove poli­cies to max­imise pos­i­tive out­comes and re­duce risks, such as the pos­si­bil­ity of low rates of adop­tion or lim­ited in­sti­tu­tional ca­pac­ity for ef­fec­tive im­ple­men­ta­tion.

It is not enough sim­ply to as­sume that the data revo­lu­tion will ben­e­fit sus­tain­able de­vel­op­ment. En­sur­ing that it does will re­quire recog­nis­ing the im­por­tance of rig­or­ous anal­y­sis in ev­ery data-col­lec­tion ef­fort and the for­ma­tion of a new gen­er­a­tion of de­ci­sion sci­en­tists to work along­side pol­i­cy­mak­ers.

Newspapers in English

Newspapers from Cyprus

© PressReader. All rights reserved.