Suppose that you discover a colleague has run a randomized trial in a particularly poor area...

 

Transcribed Image Text:

Suppose that you discover a colleague has run a randomized trial in a particularly poor area of rural Nairobi. In this setting, all farmers initially reported having no access to credit (i.e., their measure of Credit was equal to 0), but your colleague then randomly provided concessional loan to half of the farmers through a local bank so that they now had access to credit. After implementing this intervention, your colleague estimates the same regression as a). However, the regression coefficient b that your colleague estimates is very different from your estimate of the coefficient b. (10 points) Summarize the argument presented in Lecture 4 regarding how randomization solves the selection bias problem. Then, given the differing b estimates, explain which estimate is more likely to reflect the causal impact of access to credit on fertilizer application rate. Does the fact that farmers randomly assigned to the concessional loan treatment subsequently report spending more time on their other jobs affect your response? Why or why not? (10 points) Provide two explanations for why the b estimates from the two studies (in Kenya and Nairobi) would not be expected to be identical even if each reflected the causal impact of access to credit on fertilizer use in the given study setting. (5 points) So far, we have focused exclusively on the size/magnitude of the b estimate, but we have not discussed its statistical significance. Explain how (if at all) the policy implications of your estimate of the causal impact of credit (b) should differ depending on whether the coefficient is statistically distinguishable from zero (i.e., whether we can reject the null hypothesis that the coefficient is equal to 0 with 95% confidence). Suppose that you discover a colleague has run a randomized trial in a particularly poor area of rural Nairobi. In this setting, all farmers initially reported having no access to credit (i.e., their measure of Credit was equal to 0), but your colleague then randomly provided concessional loan to half of the farmers through a local bank so that they now had access to credit. After implementing this intervention, your colleague estimates the same regression as a). However, the regression coefficient b that your colleague estimates is very different from your estimate of the coefficient b. (10 points) Summarize the argument presented in Lecture 4 regarding how randomization solves the selection bias problem. Then, given the differing b estimates, explain which estimate is more likely to reflect the causal impact of access to credit on fertilizer application rate. Does the fact that farmers randomly assigned to the concessional loan treatment subsequently report spending more time on their other jobs affect your response? Why or why not? (10 points) Provide two explanations for why the b estimates from the two studies (in Kenya and Nairobi) would not be expected to be identical even if each reflected the causal impact of access to credit on fertilizer use in the given study setting. (5 points) So far, we have focused exclusively on the size/magnitude of the b estimate, but we have not discussed its statistical significance. Explain how (if at all) the policy implications of your estimate of the causal impact of credit (b) should differ depending on whether the coefficient is statistically distinguishable from zero (i.e., whether we can reject the null hypothesis that the coefficient is equal to 0 with 95% confidence).

Answer rating: 100% (QA)

Answer d Randomization solves the selection bias problem by randomly assigning individuals to the treatment or control group thereby ensuring that the ... View the full answer