![Sample our Social Sciences journals, sign in here to start your FREE access for 14 days]({"type":"image" , "src" : "/sda/110822/TOC-Subject-Sample-2021-Social-Sciences.jpg"})
Abstract
Although a large literature highlights the impact of personality traits on key labour market outcomes, evidence of their impact on agricultural production decisions remains limited. Data from 1,200 Ghanaian rice farmers suggest that noncognitive skills (polychronicity, work centrality, and optimism) significantly affect simple adoption decisions, returns from adoption, and technical efficiency in rice production, and that the size of the estimated impacts exceeds that of traditional human capital measures. Greater focus on personality traits relative to cognitive skills may help accelerate innovation diffusion in the short term, and help farmers to respond flexibly to new opportunities and risks in the longer term.
Acknowledgments
We are grateful to Iggy Bassi, Toks Abimbola, and Karan Chopra in GADCO’s management team for their support to this research, Julius Ameku, Sridhar Reddy, and Vishnuvardhan Banda for field implementation, Markus Goldstein, Tricia Koroknay-Palicz, Angeli Kirk, Courtney Han, Francesca Viola, and Junwei Chen at the World Bank for comments and contributions to the data collection. We thank the IPA Ghana team for excellent fieldwork, particularly Gabriel Lawin, Dziwornu Kwami Adanu, Mona Niina Iddrisu, Virginia Ceretti, Cornelius Owusu Adjei, Maham Farhat, Nicolas Marin, James Svenstrup, Philip Amara, Pace Philips, and Loic Watine. We appreciate cooperation and support from the Ghana Irrigation Development Authority (GIDA). Financial support from a USAID Trust Fund to support impact evaluation of initiatives under the GCAP project, and the Africa region PSIA Trust Fund is gratefully acknowledged. The analysis data and code can be made available to bona fide researchers on request.
Disclosure statement
No potential conflict of interest was reported by the authors.
SUPPLEMENTAL DATA
Supplementary Materials are available for this article which can be accessed via the online version of this journal available at https://doi.org/10.1080/00220388.2019.1666978
Notes
1. Throughout the paper, we refer to ‘personality traits’ as comprising both personality and motivation traits.
2. All female farmers were included to allow sufficient statistical power for analysis by gender sub-group.
3. Of the 1,600 farmers in the sample, 159 were dropped because they did not cultivate irrigated rice parcels, 80 because accurate harvest information was not available, and another 184 due to missing control variables.
4. Transplanting is less widespread in the Weta than in the Kpong irrigation scheme (3% vs 49%). , and therefore compare farmers who transplant with those who do not within Kpong only in columns 4 and 5 to avoid conflating differences between transplanters and non-transplanters with differences between Weta and Kpong.
5. Cultivators located in Weta apply 30 per cent less NPK-Activia per hectare than those in Kpong, but offset this with higher use of Urea and Ammonia, resulting in comparable levels of N application. There is no statistically significant difference in application of nutrients between male and female cultivators.
6. They have been shown to predict innovation and success for micro entrepreneurs in Sri Lanka (De Mel et al., Citation2009). Having the same set of questions allows us to test if traits that predicted micro-entrepreneurial success in Sri Lanka are relevant for smallholder productivity and technology adoption in Ghana, implying some transferability of personality traits across activities and cultural contexts. The procedure is identical to that in De Mel et al. (Citation2010), except that the construct values there are not divided by the number of questions comprising the category.
7. They find a test–retest correlation of 0.73 when non-cognitive skill questions were administered at the end of the interview, just above the minimum of 0.7 often applied to define reliability.
8. Acquiescence bias is the tendency of some respondents to agree with the contents of a question regardless of the content. Since the non-cognitive skill response categories include an unequal number of positively and negatively phrased questions, the trait indicator values may reflect both the underlying trait and the extent of the respondent’s acquiescence bias (Laajaj & Macours, Citation2017; Rammstedt & Farmer, Citation2013; Soto et al., Citation2008).
9. Due to the structure of our instrument, similar positively and negatively phrased questions were not available for each trait category. The question pairs used for the correction are identified in the Supplementary Materials.
10. P-value < 0.0001.
11. P-value = 0.1629.
12. While some psychometric studies have also applied ipsatization (dividing by the standard deviation of all responses after subtracting the mean), this practice is debated in psychometric literature, and was found by (Laajaj & Macours, Citation2017) to worsen the validity and reliability of non-cognitive constructs. We do not apply it here.
13. While the literature suggests that the factor analysis deriving the Big Five Factor structure can be replicated across cultures (McCrae & Costa, Citation1997), personality indicators for individuals may vary systematically across areas due to environmental or cultural characteristics (see, McCrae & Terracciano, Citation2005).
14. This could also be understood as a reversed scale of the perceived cost of adoption.
15. This is also consistent with recent theoretical and empirical research highlighting the importance of internal constraints such as perceived self-efficacy (Wuepper & Lybbert, Citation2017) and hope (Lybbert & Wydick, Citation2018) to recognising investment opportunities, with important implications for poverty and economic development.
16. The maximum likelihood estimation is performed in STATA 13 using the sfcross command (Belotti, Daidone, Ilardi, & Atella, Citation2012). Likelihood ratio tests are used to test for significance of the Z terms.
17. Psychology literature on farmers’ stress consistently shows time management to be a leading factor (Alpass et al., Citation2004; J. Deary, Willock, & McGregor, Citation1997; McGregor, Willock, & Deary, Citation1995; Pollock, Deaville, Gilman, & Willock, Citation2002; Walker & Walker, Citation1987).
18. VIF is a measure of how much the variance of the predictor’s coefficient is increased by the inclusion of other predictors in the model. Hair, Anderson, Tatham, and Black (Citation1995) suggest that a VIF below 10 indicates inconsequential collinearity.
19. Forward stepwise regression is a standard model selection algorithm that, starting with no variables in the model, sequentially adds the predictor most correlated with the residual of the outcome variable after all variables currently in the model are controlled for until no predictor outside of the model meets a minimum p-value when added to the model (in this case, 0.09). Predictors inside the model that no longer meet a minimum p-value requirement (in this case 0.1) as a result of the addition of other variables to the model are then removed until all predictors in the model meet the minimum significance requirement. The process is then repeated until all predictors inside and outside the model are above or below their minimum significance requirement, respectively. For the stochastic frontier selection inclusion of all production function variables is forced, while the algorithm selects-in covariates from the conditional mean.
20. Least angle regression (LARS) is similar to a forward stepwise procedure but avoids arbitrarily removing predictors highly correlated with the outcome variable that happen to be correlated with another predictor selected earlier. It does so by increasing the coefficients on covariates currently in the model in their joint least squares direction until a variable not currently in the model has as high a correlation with the residual as the variables currently in the model. At this point, that variable is added and the procedure is repeated. LARs achieves the same result as the LASSO selection technique (Tibshirani, Citation1996) except when the coefficient of a variable already in the model hits zero, in which case it is removed and the joint direction is recomputed (Efron et al., Citation2004).
21. To test for robustness, we also remove traits with Cronbach’s alpha scores below 0.4 and repeat the above procedure. The Cronbach’s alpha scores range from 0.41 to 0.61 for the unadjusted traits, and 0.61–0.81 for the acquiescence bias adjusted traits with the exception of impulsiveness and locus of control, which are removed for the robustness check.
22. These results are unaffected by either adjustment for acquiescence bias (Supplementary Material Table 1), or removal of traits with a low Cronbach’s alpha score (Supplementary Material Table 3).
23. Results for mechanised harvest adoption are available from the authors on request.
24. To allow consistent interpretation, values for traits with negative coefficients are re-scaled so that higher deciles increase the probability of adoption. In particular, the indicators for cultivator achievement motivation, power motivation, work centrality/passion and age are reversed.
25. The three panels of display the predicted probability of adopting transplanting at each decile of the distribution of (i) education, age and experience as traditional human capital variables (panel A); (ii) all non-cognitive traits included in (panel B); and (iii) the digitspan as a possibly more accurate measure of cognitive ability (panel C). All other covariates are held constant at their sample means and the 95 per cent confidence band is computed using the delta method.
26. The increase in predicted probability of adoption is driven almost entirely by education, rather than age or experience.
27. The Cobb-Douglas is rejected in favour of the translog at the 1 per cent level for specifications both with and without non-cognitive traits.
28. When estimated on the acquiescence bias adjusted traits (Supplementary Material Table 2), the coefficient on work centrality/passion reduces by a third and becomes insignificant in some specifications, while the coefficients on power motivation, organisation and optimism increase and become significant in some specifications. Removal of traits with a low Cronbach’s alpha score (Supplementary Material Table 4) does not affect the results.