--- In [email protected], "authfriend" <[EMAIL PROTECTED]> > Data from the past several years (not positive it was > five years; it may have been fewer) was used to "predict" > what the crime rate *would* have been had the intervention > not taken place. The eight-week period for the year of > the intervention would have had buffers of a few weeks > fore and aft.
Thats where it starts to get funky. In both regression and ARMIA models, regression is better suited, a properly specified model will control for, and estimate the effects of, all the relevant effects, including the intervention. (And via differencing and lag variables, regression can adjust for all the seasonal and autocorrelation effects that ARIMA does) Simplifying things a bit, a regression model will estimate, "coeficients" aka betas, for all the included independent variables in the form of a specified linear equation. Thus, as a simple example, the estimated betas and equations would be of the form: Crime = (b1 x weather) + (b2 x # of police) + (b3 x seasonal variable) + (b4 x ME intervention) + e b= betas e = error term The model can be used to "predict" or indicate what the crime rate would be for different levels or the absence of any of the independent variables. For example, using the ME var as observed in the study, the equation shows approximately what the observed crime rate was. By setting the ME var to zero, one can see what the crime rate would have been without the intervention. Or the weather var can be set to average temperatures, to see how that would have lessened crime, compared to the high, hot temperatures actually observed.* This is the power of regression analysis. It controls for each independent variable, and allows predictions for the absence or adjustment of any or all variables. Lots of "what-if" questions can be addressed. So when separate models are used to predict crime, distinct from the intervention model, its unconventional. Its not needed, unless there were severe data problems. Doing so weakens the predictive power of the model(s). There is nothing in the data issues, at first glance, that suggest why multiple, models were used. And doing so, it can create complexity, raises all sorts of problems, and can weaken the ability to make use the strong statistical statements that can be made with properly constructed regression models: such as "85 % of the variations in crime are explained by the model" and "each independent variable was significant to such and such a level" and there is low correlation between independnet variables", and "there is low autocorrelation in the model, etc." And such disparate models "glued togeteher" are more prone to fudging (researcher cherry-picking) and general error. So it sounds funky. Its not clear what they did this separate model approach. If thats what they indeed did. Its not standard. But without seeing the study, its hard to say what they did and why. ------ *(ARIMA does a similar thing, but is used primarily when the data is highly "auto correlated" -- that is past values t1, or t2 are correleated with current t0 values. Or is highly seasonal. This is typically economic and financial data. "Social" data typically does not have high autocorrelations. Thus, why ARIMA was used is not clear. Any seasonal effects can be well handled in regression) ------------------------ Yahoo! Groups Sponsor --------------------~--> Get fast access to your favorite Yahoo! Groups. Make Yahoo! your home page http://us.click.yahoo.com/dpRU5A/wUILAA/yQLSAA/JjtolB/TM --------------------------------------------------------------------~-> To subscribe, send a message to: [EMAIL PROTECTED] Or go to: http://groups.yahoo.com/group/FairfieldLife/ and click 'Join This Group!' Yahoo! Groups Links <*> To visit your group on the web, go to: http://groups.yahoo.com/group/FairfieldLife/ <*> To unsubscribe from this group, send an email to: [EMAIL PROTECTED] <*> Your use of Yahoo! Groups is subject to: http://docs.yahoo.com/info/terms/
