National Data Coordinators (NDCs) were trained in the project methodology. They were required to identify up to eight other road safety experts within their country from different sectors and to facilitate a consensus meeting of these respondents. While each expert responded to the questionnaire in their individual capacity, the consensus meeting facilitated by NDCs allowed for discussion of all responses, and the group used this discussion to agree one final set of information that best represented their country?s situation at the time (up to 2011, using the most recent data available). This was then submitted to the World Health Organization (WHO), see Figure E1.
Data collection began in May 2011 and was completed by December 2011. Validation involved checking data for logical inconsistencies, and these were checked with National Data Coordinators. Following the validation process, final data sets were sent to respective governments for review and sign-off.
Following the computation of estimates of road traffic deaths for 2010, a country consultation process was undertaken. Each country was provided with an opportunity to comment on both the methodology which had been employed to compute the estimate, as well as the actual estimate received. As a result of this process, seven countries (Canada, Chile, China (14), Costa Rica, India (15), Iran and the USA) provided WHO with more up to date data which was used to improve estimates.
These are quotes from pages 42, 49 and 50 of the report - basically a group of Dominican road safety experts got together and compiled and agreed to the stats for the country, then the DR government was allowed to provide more data to improve the estimates if the government desired or to ask questions about the methodology of the study, which it didn't btw.
The methodology and safeguards to ensure accuracy seem quite adequate to me (and to the government of the DR evidently) - to which modifier of the real statistics are you referring to in your comment Chip? I am just curious - not looking to pick a fight.
"
Group 4: Countries without eligible death registration data
For 78 countries that did not fall into Groups 1, 2 or 3, a regression model was used to estimate total road traffic deaths. As in the first report, a negative binomial regression model was used ? appropriate for modelling non-negative integer count data (number of road traffic deaths) (7, 8). A likelihood ratio test was used to assess that the negative binomial model provided a better fit to the data than a Poisson model (where the variance of the data is constrained to equal the mean).
ln N = C +β1 X1 +β2X2 +....+βnXn + ln Pop +ε
where N is the total road traffic deaths (for a country-year), C is a constant term, X1 are a set of explanatory covariates, Pop is the population for the country-year, and ε is the negative binomial error term. Population was used as exposure, making it possible to interpret the coefficients (1) for the independent variables as effects on rates rather than a count. In a previous study, this type of model was used to represent ?accident proneness?(9). Other authors have also found a negative binomial regression model to be the
appropriate for count data such as road traffic fatalities (10).
The parameters β1, β2 ??? βn (in the equation above) were
estimated by fitting the negative binomial regression model to estimated total road traffic deaths from death registration data for all country-years in the range 1950-2010 meeting the completeness criteria (Group 1).
Three models (Models A, B and C) were chosen that had good in-sample and out-of-sample fit, and for which all the covariates were statistically significant and for which overall estimation is the average of the prediction of these three best models (see Table E2). For these countries a 95% confidence interval was given by using the negative binomial regression in the statistical package STATA."