Impact Assessment

[Bitwise]

Hi,

I'm performing an interrupted time series (Box-Jenkins) design to assess whether anti-spam legislation has impacted email spam rates. I've done fine with the identification, estimation, and diagnosis portions of this process, and the data is ready for impact assessment. However, I can't figure out how to estimate the two parameters I need to assess the intervention. I've created a dummy variable for 0 = preintervention and 1 = postintervention. I've decided my model should use a gradual, permanent impact. Now I need, according to what I've read, to estimate two coefficients, omega and delta (yt = ƍyt-1 + ωIt), so that I can interpret whether there was a significant impact. So how do I estimate these two parameters?

[startz]

Code: Select all

ls y c y(t-1) I

or maybe

Code: Select all

ls y c y(t-1) I ar(1) ar(2) etc.

[Bitwise]

Thanks for your help. I tried the equation you suggested (I have an ARIMA(2,0,0) model). I used my dummy variable (0 = preintervention, 1 = postintervention) as “I” (named “impact”) and my time series variable as “y” (named “hydepark”). I entered it as this:

Code: Select all

hydepark c hydepark(-1) impact ar(1) ar(2)

And got the following output:

Code: Select all

Variable Coefficient Std. Error t-Statistic Prob. C 4.523364 2.133556 2.120106 0.0379 HYDEPARK(-1) 0.738460 0.120748 6.115715 0.0000 IMPACT -2.020163 1.231077 -1.640972 0.1058 AR(1) -0.484213 0.174125 -2.780841 0.0071 AR(2) -0.008724 0.165208 -0.052804 0.9581

However, this data I’m analyzing here is from a book with step by step instructions on ITSA. I’ve been able to match all the output in Eviews 6 that the book has in the instructions, except for the impact assessment phase. The output I got above doesn’t quite match what the book has yet. What I need are those two parameters (omega and delta) to be estimated, and then I can plug them into the following forumula: yt = ƍyt-1 + ωIt to determine what kind of impact I have.

The book’s not clear on how to estimate these two parameters.

Thanks,
-Bitwise

[startz]

Based on your EViews estimates ƍ=0.738460 and ω=-2.020163

[Bitwise]

That's strange because the book I'm reading says the estimates should be ar(1) = .242, ar(2) = .336, δ = .927, and ω = -.781. Perhaps the books is using something other than least squares? Everything's matched the book up until now, but I suppose I could have entered something wrong at one step or another.

Thanks,
-Bitwise