Skip to main content



Child-Maltreatment-Research-L (CMRL) List Serve

Browse or Search All Past CMRL Messages

Welcome to the database of past Child-Maltreatment-Research-L (CMRL) list serve messages (10,000+). The table below contains all past CMRL messages (text only, no attachments) from Nov. 20, 1996 - June 11, 2018 and is updated quarterly.

Instructions: Postings are listed for browsing with the newest messages first. Click on the linked ID number to see a message. You can search the author, subject, message ID, and message content fields by entering your criteria into this search box:

Message ID: 8227
Date: 2009-08-03

Author:Chaffin, Mark J. (HSC)

Subject:RE: Rise in CM fatalities and classification of neglectful drownings

David,



I think the issues people are raising about alternative explanations for the observed increases are excellent, and clearly call out for some examination. I would add a couple of additional thoughts on the issue of how to interpret a sudden increase in maltreatment fatality rates in some states. First, looking across states, it might be good to check for slope-intercept correlations in change over time data. That is, are the recent increases seen more often in states that tended to have low rates to begin with? Second, when looking at cross-state data, I've found that maltreatment fatality rates tend to be highly correlated with other things (e.g. state rates of violent crime, poverty, teen births, child welfare reports, etc.). As a simple multiple regression, you can easily account for over half the variance with these predictors, and there also are clear outliers from the multivariate prediction line (not just normal variation from a prediction line, but huge outliers), especially if you exclude the outlier state from prediction model development. So, to expand the question, one might examine not only slope-intercept covariance, but also covariance between slopes and prediction residuals. And if changes in maltreatment fatality parallel changes in known predictors and risk factors. This might shed some light on whether the change essentially represents some sort of "correction" (i.e. procedural changes, regression toward the mean, or changes in random error), or represents a genuine change. Checking the time series of rates for individual states, over time, using some sort of autoregressive or moving average structure might also be informative for checking that the observed increases are actually different from random variation.



Mark





David,



I think the issues people are raising about alternative explanations for the observed increases are excellent, and clearly call out for some examination. I would add a couple of additional thoughts on the issue of how to interpret a sudden increase in maltreatment fatality rates in some states. First, looking across states, it might be good to check for slope-intercept correlations in change over time data. That is, are the recent increases seen more often in states that tended to have low rates to begin with? Second, when looking at cross-state data, I've found that maltreatment fatality rates tend to be highly correlated with other things (e.g. state rates of violent crime, poverty, teen births, child welfare reports, etc.). As a simple multiple regression, you can easily account for over half the variance with these predictors, and there also are clear outliers from the multivariate prediction line (not just normal variation from a prediction line, but huge outliers), especially if you exclude the outlier state from prediction model development. So, to expand the question, one might examine not only slope-intercept covariance, but also covariance between slopes and prediction residuals. And if changes in maltreatment fatality parallel changes in known predictors and risk factors. This might shed some light on whether the change essentially represents some sort of "correction" (i.e. procedural changes, regression toward the mean, or changes in random error), or represents a genuine change. Checking the time series of rates for individual states, over time, using some sort of autoregressive or moving average structure might also be informative for checking that the observed increases are actually different from random variation.



Mark