Search

Single-centre studies examining the transgenerational inheritance of pathologies in rodents exposed to pesticides have not always taken important design and analysis issues into account. This paper examines these methodological and statistical issues in detail. Its particular focus is on the estimation of ‘litter effects’: the tendency for rodents within a litter to be more alike than rodents in different litters. Appropriate statistical models were fitted to publisheddata from a series of widely reported studies carried out at Washington State University. These studies were amalgamated into a single dataset in order to estimate these litter effects and associated treatment effects. Litter effects varied by outcome and were often substantial. Consequently, the effective sample size was often substantially less than the number of observations with implications for the power of the studies. Moreover, the reported precision of the estimates of treatment effects was too low. These problems are exacerbated by unexplained missing data across generations. Researchers in the life sciencescould be more cognisant of the guidelines established in medicine for reporting randomised controlled trials, particularly cluster randomised trials. More attention should be paid to the design and analysis of multi-generational rodent studies; their imperfections have important implications for assessments of the evidence relating to the risks of pesticides for public health.

More than a decade has passed since Bt cotton was introduced in India. It is now possible to use official cotton statistics to assess whether Bt cotton has had positive effects on farmers’ lives and livelihoods. I use interrupted time-series analysis of data on insecticide costs, yields, and profits to examine trends before and after the introduction of Bt cotton in the States of Haryana, Punjab, and Rajasthan, in northern India. The conclusions from these analyses are mixed. The more expensive Bt hybrid seeds have lowered insecticide costs in all three States, but only in Rajasthan did yields increase. An important message of this paper is that conclusions about the effectiveness of Bt cotton are more nuanced than many researchers and commentators recognise. The paper does not refute the assertions about the success of Bt cotton, but it does show that the benefits are not evenly distributed across India.

The evidence for transgenerational transmission of pathologies in rodents exposed to pesticides is largely based on studies with weaknesses in design and analysis and a lack of transparency in reporting the results. This paper examines these methodological and statistical issues in detail. Its particular focus is on the estimation of ‘litter effects’: the tendency for rodents within a litter to be more alike than rodents in different litters. Appropriate statistical models were fitted to published data from a series of widely reported studies carried out at Washington State University. These studies were amalgamated into a single dataset in order to estimate these litter effects and the associated treatment effects. Litter effects varied by outcome and were often substantial. Consequently, the effective sample size was often substantially less than the number of observations with implications for the power of the studies. Moreover, the reported precision of the estimates of treatment effects was too low. These problems are exacerbated by unexplained missing data across generations. Researchers in the life sciences could be more cognisant of the guidelines established in medicine for reporting randomised controlled trials, particularly cluster randomised trials. More attention should be paid to the design and analysis of multi-generational rodent studies; their imperfections have important implications for assessments of the evidence relating to the risks of pesticides for public health.

Interviewers made four observations related to future participation, respondent cooperation, enjoyment and whether the respondent found the questions difficult, for a large sample of face-to-face interviews at wave four of the UK Millennium Cohort Study (MCS). The focus of the paper is on predicting response behavior in the subsequent wave of MCS, four years later. The two most predictive observations are whether the respondent is likely to participate in the next wave and whether they enjoyed the interview. Not only do these predict non-response at the next wave, they do so after controlling for other explanatory variables from earlier waves in a response propensity model. Consequently, these two interviewer observations improve discrimination between respondents and non-respondents at wave five as estimated by Gini coefficients generated by a Receiver Operating Characteristic curve analysis. The predicted probabilities of responding at wave five are also used to estimate R-indicators, particularly to address the question of whether, hypothetically, conversion of ‘frail’ respondents would lead to improved representativity and reduced bias in longitudinal estimates of interest. The evidence from the R-indicators and partial R-indicators suggests that successful conversions could achieve those aims although the cost of so doing might outweigh the benefits., methods, data, analyses, Vol 11, No 1 (2017)