Researchers want to estimate how much each additional kilogram of fertilizer per plot changes wheat yield. Assume that over the range of amounts considered, mean yield responds to the amount applied in an approximately linear way, so that a single slope can describe the response. In which scenario can the regression slope of yield on fertilizer amount most directly be interpreted as the average effect of fertilizer in the range studied?
Sixty similar plots are each randomly assigned one of six fertilizer amounts from 0 to 10 kg; each plot receives its assigned amount, and yields are regressed on the amounts actually applied.
Observational farm data are analysed with a multiple regression that includes soil pH and rainfall as additional predictors.
Data from 300 observational plots give r2 = 0.92 for yield on fertilizer.
Records from 5,000 farms show how much fertilizer each farmer chose to apply and the resulting yield; the regression slope is highly significant.
Unlock full course notes by signing up
Objectives of Regression (Prediction, not Causality): difficulty 4
For 52 weeks at a retail chain, weekly advertising spend x (thousands of dollars) has mean 12 and standard deviation 3, weekly sales y (thousands of units) has mean 50 and standard deviation 8, and the correlation is r = 0.75. Across these 52 weeks advertising spend ranged from 5 to 19 thousand dollars and sales ranged from 32 to 68 thousand units, so the advertising value 15 and the sales value 58 used below both lie inside the observed ranges.
Find the slope of the least-squares line for predicting sales from advertising, and use it to predict sales in a week with advertising spend 15.
Find the slope of the least-squares line for predicting advertising from sales, and use it to predict advertising spend in a week with sales of 58. Also find the advertising value obtained by solving the line from part 1 for x when y = 58.