RockTheVote               package:pscl               R Documentation

_V_o_t_e_r _t_u_r_n_o_u_t _e_x_p_e_r_i_m_e_n_t, _u_s_i_n_g _R_o_c_k _T_h_e _V_o_t_e _a_d_s

_D_e_s_c_r_i_p_t_i_o_n:

     Voter turnout data spanning 85 cable TV systems, randomly
     allocated to a voter mobilization experiment targetting 18-19 year
     olds with "Rock the Vote" television advertisments

_U_s_a_g_e:

     data(RockTheVote)

_F_o_r_m_a_t:

     A data frame with 85 observations on the following 6 variables.

     '_s_t_r_a_t_a' numeric, experimental strata

     '_t_r_e_a_t_e_d' numeric, 1 if a treated cable system, 0 otherwise

     '_r' numeric, number of 18 and 19 year olds turning out

     '_n' numeric, number of 19 and 19 year olds registered

     '_p' numeric, proportion of 18 and 19 year olds turning out

     '_t_r_e_a_t_e_d_I_n_d_e_x' numeric, a counter indexing the 42 treated units

_D_e_t_a_i_l_s:

     Green and Vavreck (2008) implemented a cluster-randomized
     experimental design in assessing the effects of a voter
     mobilization treatment in the 2004 U.S.~~Presidential election. 
     The clusters in this design are geographic areas served by a
     single cable television system. So as to facilitate analysis, the
     researchers restricted their attention to small cable systems
     whose reach is limited to a single zip code.  Further, since the
     experiment was fielded during the last week of the presidential
     election, the researchers restricted their search to cable systems
     that were not in the 16 hotly-contested ``battleground'' states
     (as designated by the _Los Angeles Times_).

     Eighty-five cable systems were available for randomization and
     were assigned to treatment after stratification on previous
     turnout levels in presidential elections (as determined from
     analysis of the corresponding states' voter registration files). 
     Each cable system was matched with one or sometimes two other
     cable systems in the same state, yielding 40 strata.  Then within
     each strata, cable systems were randomly assigned to treatment and
     control conditions.  Strata 3, 8 and 25 have two control cable
     systems and 1 treated system each, while strata 6 and 20 have two
     treated cable systems and one control system. The remaining 35
     strata have 1 treated cable system and 1 control system.  In this
     way there are 38 + 4 = 42 treated systems, spanning 40 experiment
     strata.

     The treatment involved researchers purchasing prime-time
     advertising spots on four channels in the respective cable system
     in which the researchers aired voter mobilization ads.  The ads
     were produced by _Rock the Vote_, targeted at younger voters, and
     aired four times per night, per channel, over the last eight days
     of the election campaign.  After the election, public records were
     consulted to assemble data on turnout levels in the treated and
     control cable systems.  In the analysis reported in Green and
     Vavreck (2008), the researchers focused on turnout among
     registered voters aged 18 and 19 years old.

_R_e_f_e_r_e_n_c_e_s:

     Green, Donald P. and Lynn Vavreck. 2008. "Analysis of
     Cluster-Randomized Experiments: A Comparison of Alternative
     Estimation Approaches." _Political Analysis_ 16:138-152.

_E_x_a_m_p_l_e_s:

     data(RockTheVote)
     ## estimate MLEs of treatment effects
     deltaFunction <- function(data){
       model <- glm(cbind(r,n-r)~treated,
                    data=data,
                    family=binomial)
       c(coef(model)[2],
         confint(model)[2,])
     }

     tmp <- by(RockTheVote,
               as.factor(RockTheVote$strata),
               deltaFunction)

     tmp <- matrix(unlist(tmp),ncol=3,byrow=TRUE)

     indx <- order(tmp[,1])

     plot(y=1:40,
          x=tmp[indx,1],
          pch=16,cex=1.25,
          xlim=range(tmp),
          ylab="",
          axes=FALSE,
          xlab="Estimated Treatment Effect (MLEs, Logit Scale)")
     text(y=1:40,
          x=par()$usr[1],
          pos=4,
          as.character((1:40)[indx]),
          cex=.5)
     segments(x0=tmp[indx,2],
              x1=tmp[indx,3],
              y0=1:40,
              y1=1:40)
     axis(1)
     axis(3)
     abline(v=0)

