Measuring Skills in Developing Countries

Notes: NA is not applicable; Cronbach’s alpha cannot be calculated when there is only one item. IRT is item response theory. In Panel B noncognitive variables have been demeaned to correct for the acquiescence bias.

Notes: Dependent variable is the average rank of maize yields calculated over the four seasons (short rain 14 to long rain 16). Controls include education, literacy, gender, age and age squared of the farmer, land and cattle ownership, size and quality of the house, household size, whether the farmer is the household head, household head’s gender, village fixed effects, and enumerator-assignment fixed effects. Standard errors clustered at village level in parentheses. Significance: *p < 0.1, **p < 0.05, ***p < 0.01.

Notes: Dependent variable is the average rank of maize yields calculated over the four seasons (short rain 14 to long rain 16). Controls include education, literacy, gender, age and age squared of the farmer, land and cattle ownership, size and quality of the house, household size, whether the farmer is the household head, household head’s gender, village fixed effects, and enumerator-assignment fixed effects. NC is noncognitive. Standard errors clustered at village level in parentheses. Significance: *p < 0.1, **p < 0.05, ***p < 0.01.

Notes: Dependent variable is the average rank of maize yields calculated over the four seasons (short rain 14 to long rain 16). Controls include education, literacy, gender, age and age squared of the farmer, land and cattle ownership, size and quality of the house, household size, whether the farmer is the household head, household head’s gender, village fixed effects, and enumerator-assignment fixed effects. IRT is item response theory. LOC is locus of control. NC is noncognitive. Standard errors clustered at village level in parentheses. Significance: *p < 0.1, **p < 0.05, ***p < 0.01.

Notes: Dependent variables are the averages of binary variables calculated over the four seasons (short rain 14 to long rain 16). Controls include education, literacy, gender, age and age squared of the farmer, land and cattle ownership, size and quality of the house, household size, whether the farmer is the household head, household head’s gender, village fixed effects, and enumerator-assignment fixed effects. Standard errors clustered at village level in parentheses. Significance: *p < 0.1, **p < 0.05, ***p < 0.01.

Notes: Dependent variables are the averages of binary variables calculated over the four seasons (short rain 14 to long rain 16). Controls include education, literacy, gender, age and age squared of the farmer, land and cattle ownership, size and quality of the house, household size, whether the farmer is the household head, household head’s gender, village fixed effects, and enumerator-assignment fixed effects. IRT is item response theory. NC is noncognitive. Standard errors clustered at village level in parentheses. *p < 0.1, **p < 0.05, ***p < 0.01.

Notes: The p-values between parentheses indicate significance of the difference between same and different enumerator assigned for test and retest. R² of enumerator fixed effects (FE) is the R² of a regression of the improved construct on (randomly) assigned enumerator fixed effects. The p-values between brackets provide the significance levels of the R² values. All p-values were obtained by randomization inference (10,000 repetitions).

Notes: Table reports correlations between the 14 summary questions and the subconstruct most closely corresponding to each question. We use the demeaned measures of noncognitive subconstructs and the improved indexes of cognitive subconstructs and technical skills.

Notes: Skill proxies obtained through village informant (CHW), scored on scale from 1 (low) to 3 (high). The right side of the table presents the correlation of the three questions asked to the village informant with the improved index of the corresponding skill, which the question intended to proxy. In the regressions, the dependent variable is the average rank of maize yields calculated over the four seasons (short rain 14 to long rain 16). Controls include education, literacy, gender, age and age squared of the farmer, land and cattle ownership, size and quality of the house, household size, whether the farmer is the household head, household head’s gender, village fixed effects (Vil. FE), and enumerator-assignment fixed effects. Standard errors clustered at village level in parentheses. Significance: *p < 0.1, **p < 0.05, ***p < 0.01.

Notes: To have comparable results for different countries, the same subset of questions was kept, using Big Five data from other countries, and the factorial analysis uses varimax rotation. The absolute value of correlations between factor loads is used to calculate congruence. To maintain comparability the items are not corrected for acquiescence bias.