Page 53 - Contributed Paper Session (CPS) - Volume 1

CPS863 Abubakar S. A. et al.
2.4.2 Model 2
A mixed effect model with total corn area of previous quarter as fixed
e↵ects and municipal random effect. = + −1 + + , ∽

2
(0, ), −1 ≡ corn area production at municipality at quarter − 1 and
~(0, ).

2.4.3 Model 3
A mixed effect model with municipal random effect and total corn area and
published provincial yields as fixed effect. = + −1 ∗ +
+ ,

2
∽ (0, ), ≡ provincial average yield at municipality at quarter −

1 and ~(0, ). Each model was developed using CPS survey data collected

from years 2013 to 2015 and bootstraps estimates of production with 200
replications were computed.

3. Results
3.1 Determination of Initial Number of Sample Barangay
For the first phase of this research study, the initial sample barangays was
determined. Corn farm area was used as the stratification variable when using
the LH algorithm since it is highly correlated with corn production. Among all
combinations of CVs (1%, 5% and 8%) and strata (3, 5, and 10), five (5) strata
and 5% CV yielded the optimal sample size. Results revealed that using 10
strata would result to too many zero cells while using three (3) strata on the
other hand would result to too large sample size per strata. For the CVs, the
sample size would be too small if 8% CV was used while using 1% CV would
result to too large sample size. The total computed initial number of barangays
is 1,952.

3.2 Determination of the Best Sampling Design
For the second phase of this study, the di↵erent scenarios for the
determination of best sampling design was performed using 5% CV and five
(5) strata. Results showed that stratified-LH is the best stratification in terms
of precision and accuracy with PPS Systematic as the best sampling selection
method for sample barangays, in general. Also, SRSWOR is the best selection
method for sample households. Further, it was observed that accuracy and
precision generally increase as the number of sample household increases.
Hence, of the three (3) sample sizes considered (10, 15 and 20) households), a
sample of 20 households is the best in terms of CVs and MAPEs. Also, the CVs
and MAPEs are high for minor producing provinces/HUCs.

3.3 Determination of the Best Estimation Procedure
Results for the third phase showed that modified bootstrap 2nd stage has
the lowest trimed CVs in all quarters with CVs ranging from 4.8% to 14.3%,

42 | I S I W S C 2 0 1 9