Decentralization and Corruption in the Philippines
Decentralization and Corruption in the Philippines
Decentralization and Corruption in the Philippines
Create successful ePaper yourself
Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.
perceptions of corruption us<strong>in</strong>g answers to specific questions about different corrupt<br />
practices <strong>in</strong> <strong>the</strong> health <strong>and</strong> adm<strong>in</strong>istration office, cleaned of perception bias us<strong>in</strong>g<br />
equations 5.1 <strong>and</strong> 5.2, <strong>and</strong> <strong>the</strong>n comb<strong>in</strong>ed <strong>in</strong>to one <strong>in</strong>dex. We present results us<strong>in</strong>g each<br />
<strong>in</strong>dex because it is simplest to <strong>in</strong>terpret <strong>the</strong> impact of an <strong>in</strong>dependent variable on a<br />
dependant variable when <strong>the</strong> two variables are based on <strong>the</strong> answers of different<br />
respondents. However, if both <strong>in</strong>dexes are noisy measures of corruption, we could create<br />
a more precise measure of corruption by merg<strong>in</strong>g <strong>the</strong> two <strong>in</strong>dices. This does raise<br />
concerns about respondent bias, as <strong>the</strong> dependant variable would <strong>the</strong>n share a source with<br />
all <strong>in</strong>dependent variables, but because we have taken care to clean <strong>the</strong> data of perception<br />
biases we th<strong>in</strong>k <strong>the</strong>se results are of some worth.<br />
The regressions <strong>in</strong> Table 6 use municipal level data on <strong>the</strong> 80 municipalities we<br />
sampled. The first two equations 6.1 <strong>and</strong> 6.2 use household corruption perceptions as <strong>the</strong><br />
measure of corruption. The first equation is a weighted least squares estimation with <strong>the</strong><br />
weights equal to <strong>the</strong> number of observations <strong>in</strong> <strong>the</strong> municipality. This number doesn’t<br />
vary much across municipalities mak<strong>in</strong>g this equation quite similar to OLS. We quote<br />
robust st<strong>and</strong>ard errors. OLS is easy to <strong>in</strong>terpret but is probably not <strong>the</strong> appropriate<br />
estimation method because of possible prov<strong>in</strong>ce level effects. The second equation 6.2 is<br />
a r<strong>and</strong>om effects estimation that explicitly takes <strong>the</strong>se <strong>in</strong>to account.<br />
One well known hypo<strong>the</strong>sis about corruption is that when public officials enjoy<br />
more discretion, <strong>the</strong> have greater opportunities to dem<strong>and</strong> bribes, <strong>and</strong> may become more<br />
corrupt as a consequence. This idea dates back at least to Myrdal’s excellent chapter on<br />
corruption <strong>and</strong> was formalized by Klitgaard’s famous formula<br />
<strong>Corruption</strong>=Discretion – Accountability + Monopoly<br />
We beg<strong>in</strong> with <strong>the</strong> first variable- discretion. Discretion is measured by adjustability<br />
as reported by public officials <strong>and</strong> <strong>the</strong>refore is best analyzed us<strong>in</strong>g <strong>the</strong> household level<br />
data as <strong>the</strong> dependant variable. We do f<strong>in</strong>d a positive effect of adjustability on corruption<br />
but it is only significant at 20%. The coefficient of 0.18 means that a one st<strong>and</strong>ard<br />
deviation <strong>in</strong>crease <strong>in</strong> adjustability leads to an <strong>in</strong>crease <strong>in</strong> corruption of 3, or 1/6 th of a<br />
st<strong>and</strong>ard deviation. While <strong>the</strong> coefficient is not significant at conventional levels <strong>in</strong> this<br />
equation it does become significant when <strong>the</strong> comb<strong>in</strong>ed corruption <strong>in</strong>dex is used <strong>and</strong><br />
11