Table 3: Regression results for estimating slope coefficient β1
|
βo |
t-statistic |
βι |
t-statistic |
R2 | |
|
Developed countries - cutoff rate 20% | |||||
|
All rates of return ≥ 20% |
3.5859 |
43.45 |
-0.0054 |
-10.19 |
0.642 |
|
All rates of return ≥ 20% and <100% |
4.6193 |
242.54 |
-0.0257 |
-66.68 |
0.989 |
|
All rates of return ≥ 15% and <100% |
4.6340 |
315.62 |
-0.0259 |
-81.51 |
0.991 |
|
All rates of return ≥ 25% and <100% |
4.6525 |
176.98 |
-0.0262 |
-53.38 |
0.986 |
|
All rates of return ≥20% and <100%, ≤ 1985 |
3.7143 |
51.51 |
-0.0242 |
-15.98 |
0.938 |
|
All rates of return ≥20% and <100%, >1985 |
4.1117 |
188.13 |
-0.0264 |
-61.08 |
0.992 |
|
Developing countries - cutoff rate 40% | |||||
|
All rates of return ≥ 40% |
5.1689 |
81.98 |
-0.0233 |
-30.34 |
0.918 |
|
All rates of return ≥ 40% and <100% |
5.6659 |
264.60 |
-0.0308 |
-90.71 |
0.992 |
|
All rates of return ≥30% and <100% |
5.6091 |
377.08 |
-0.0300 |
-116.53 |
0.993 |
|
All rates of return ≥ 50% and <100% |
5.5628 |
196.02 |
-0.0295 |
-71.42 |
0.991 |
|
All rates of return ≥40% and <100%, ≤ 1985 |
4.3305 |
52.27 |
-0.0313 |
-25.20 |
0.975 |
|
All rates of return ≥40% and <100%, >1985 |
5.3658 |
176.03 |
-0.0305 |
-61.74 |
0.987 |
|
Africa: all rates of return ≥ 30% and <100% |
3.9918 |
52.87 |
-0.0349 |
-23.34 |
0.971 |
|
Asia: all rates of return ≥ 45% and <100% |
3.9940 |
62.36 |
-0.0162 |
-17.09 |
0.939 |
|
LAC: all rates of return ≥ 40% and <100% |
5.4419 |
80.66 |
-0.0442 |
-40.73 |
0.982 |
Differentiating the sample in studies from before and after 1985 hardly affects these results.
Given the rather approximate nature of the identified mode, extreme values for the mode were
adopted, which resulted in lower and upper bounds of the underinvestment gaps. Not surprisingly, the
underinvestment gap is quite sensitive to the estimation of the mode. Grouping the rates of return of
developing countries by region yields quite a bit of variation in both the cutoff rate and the slope
coefficient and suggests that underinvestment in agricultural R&D has been particularly high in Latin
America.
While there is considerable room for improvement in the statistical data used in the present
analysis (both in coverage as well as the quality of the rate-of-return methods used), the model cuts
quite nicely through what looked like a Gordian knot. It shows that the high means of the rate-of-
return distributions as such are no evidence of underinvestment in R&D, nor are they indicative of the
size of the underinvestment gap. The approximated modes of the rate-of-return distributions are a
substantially better indicator of underinvestment.
16