# 02 Analysis.r ### If 01 was not ran, load from jesus .g(); setScienceIfNot("EU_PricingTest_2015", load=FALSE) lib(ggplot, quie=TRUE) loadIfNotExists("DT.currency") loadIfNotExists("DT.groupings") loadIfNotExists("DT.EU_analytics") loadIfNotExists("DT.EU_analytics.bak") # loadFromJesus("DT.EU_analytics", over=TRUE) loadIfNotExists("DT.EU_aggd") # loadIfNotExists("DT.EU_aggd_reshaped.no_country") loadIfNotExists("DT.raw") get_dim_country(refresh=FALSE) setkeyIfNot(DT.country, "country_code") ggplot(DT.EU_aggd_reshaped, aes(x=preTest.units, y=Wk1.units, shape=category, color=country_code)) + geom_point(alpha=0.8, position=position_jitter(width=.05, height=.05)) + gg_xyline() + scale_y_log10() + scale_x_log10() ## Interaction plots. Two of them. par(mfrow=c(2, 2)) DT.EU_aggd[, interaction.plot(category, TestWk, units)] DT.EU_aggd[, interaction.plot(TestWk, category, units)] DT.EU_aggd[, interaction.plot(TestWk, country_code, units)] DT.EU_aggd[, interaction.plot(country_code, TestWk, units)] ## Basic models for anova model.revenue__cat_testwk <- lm(log_1p_revenue ~ category * TestWk, data=DT.EU_aggd) model.units__cat_testwk <- lm(log_1p_units ~ category * TestWk, data=DT.EU_aggd) anova(model.revenue__cat_testwk) anova(model.units__cat_testwk) ## Assumption Check par (mfrow=c(1, 2)) qqnorm(model.units__cat_testwk$res) plot(model.units__cat_testwk$fitted,model.units__cat_testwk$res,xlab="Fitted",ylab="Residuals") --------------------------------- global_alpha <- 0.2 j <- 0.15 units revenue iTunes - DT iTunes - DA Spotify - S Spotify - AS DT.EU_aggd[store_name == "Spotify"] P.iTunes_DA <- lapply(names(countries_using), function(ctry) ggplot(DT.EU_aggd[(is_iTunesDA & !is.na(TestWk))][country_code %in% ctry], alpha=global_alpha) + aes(x=TestWk.dates, y=units, color=daily_avg_album_units_30preTest, group=upc) + geom_point(alpha=global_alpha) + geom_line(alpha=global_alpha) + facet_grid("category ~ TreatmentWk1") + labs(title=paste0("Album Downloads for ", DT.country[.(ctry)]$country_name), y="Log of Units", x="") + legendtop(notitle=TRUE) + scale_y_log10(lim=c(1, 15)) + angledtext() ) P.iTunes_DT <- lapply(names(countries_using), function(ctry) ggplot(DT.EU_aggd[store_name == "iTunes" & transac_type_abbr == "DT" & album_was_purchased_incountry_during_experiment][country_code %in% ctry], alpha=global_alpha) + aes(x=TestWk.dates, y=units, color=increased_Wk1, group=upc) + geom_point(alpha=global_alpha) + geom_line(alpha=global_alpha) + facet_grid("category ~ iTunesDA_units_preTest_above5") + labs(title=paste0("Track Downloads for ", DT.country[.(ctry)]$country_name), y="Log of Units", x="") + legendtop(notitle=TRUE) + scale_y_log10() ) printToPDF(P.iTunes_DA, height = 8, width = 8 ) printToPDF(P.iTunes_DT, height = 8, width = 8 ) DT.EU_aggd[store_name == "iTunes" & transac_type_abbr == "DT"][country_code %in% ctry] --------------------------------- lims <- c(0, 30) ## count the number of units wk1 as a function of units wk 2 ggplot(DT.EU_aggd_reshaped[(is_iTunesDA)][country_code %in% ctry] [preTest.units < 30]) + aes(x=pre_preTest.units+1, y=preTest.units+1, color=TreatmentWk1, group=TreatmentWk1) + geom_point(position=position_jitter(w=j, h=j)) + # geom_line(alpha=global_alpha) + geom_smooth(alpha=global_alpha) + gg_xyline() + facet_grid("category ~ iTunesDA_units_preTest_above5") + labs(title=paste0("Album Downloads for ", DT.country[.(ctry)]$country_name), y="Log of Units", x="") + legendtop(notitle=TRUE) + ylim(lims) + xlim(lims) + log.x(lim=lims+1) + log.y(lim=lims+1) dev.new() ## count the number of units wk1 as a function of units wk 2 ggplot(DT.EU_aggd_reshaped[(is_iTunesDA & !is.na(TestWk))][country_code %in% ctry] [preTest.units < 30]) + aes(x=preTest.units+1, y=Wk1.units+1, color=TreatmentWk1, group=TreatmentWk1) + geom_point(position=position_jitter(w=j, h=j)) + # geom_line(alpha=global_alpha) + geom_smooth(alpha=global_alpha) + gg_xyline() + facet_grid("category ~ iTunesDA_units_preTest_above5") + labs(title=paste0("Album Downloads for ", DT.country[.(ctry)]$country_name), y="Log of Units", x="") + legendtop(notitle=TRUE) + ylim(lims) + xlim(lims) + log.x(lim=lims+1) + log.y(lim=lims+1) --------------------------------- cols <- setdiff(names(DT.EU_aggd_reshaped.no_country), key(DT.EU_aggd_reshaped.no_country)) DT.EU_aggd_reshaped.no_country[, (cols) := lapply(.SD, function(x) log(x + 1)), .SDcols=cols] model.ancova <- lm(Wk1.units ~ preTest.units + category - 1, data=DT.EU_aggd_reshaped.no_country) anova(model.ancova) ## Assumption Check par (mfrow=c(1, 2)) qqnorm(model.ancova$res) plot(model.ancova$fitted,model.ancova$res,xlab="Fitted",ylab="Residuals") --------------------------------- interaction.plot(DT.EU_aggd[TestWk=="preTest"]$) DT.EU_aggd[(units != (units %/% 1))] ggScatterPlot( DT.EU_analytics[store_name == "iTunes"][transac_type_abbr %in% c("DA")] [!is.na(TestWk)] , x = ) ## Poisson Check if (FALSE) { DT.EU_aggd[, list(revenue.mean = round(mean(revenue)), revenue.sd = round(sd(revenue)), units.mean = round(mean(units)), units.sd = round(sd(units))), keyby=list(store_name, transac_type_abbr, country_code)] exploreDispersion(DT.EU_aggd) } warning("Rememebr to test revenue and units", call.=FALSE) ## This week vs last week, overall P.EU_pricingtest.total_units_across_all_groups <- { title <- "Comparing total units across all groups for the same collection of UPCs before and during the experiment" ggBoxplotWithDots( # DT = DT.EU_aggd[transac_type_abbr %in% c("DT", "DA")][] DT = DT.EU_analytics[store_name == "iTunes"][transac_type_abbr %in% c("DA")] [!is.na(TestWk)] # DT = DT.EU_analytics[store_name == "iTunes"] [!is.na(TestWk)] [ upc %in% DT.EU_analytics[units > 10, upc]] , x = "TestWk" , y = "units" , dots_on = TRUE , colDotgroup = "TestWk" , dots_color = "black" , dots_alpha = 0.4 , dots_size = 0.8 , notch = TRUE , fill = "TestWk.dates" , yscale = "log" , legend="top" , legend.color=FALSE , legend_title_on = FALSE , title = chopLine(title, width=55) ) } P.EU_pricingtest_units <- { ggBoxplotWithDots( # DT = DT.EU_aggd[transac_type_abbr %in% c("DT", "DA")][] DT = DT.EU_aggd[store_name == "iTunes"][transac_type_abbr %in% c("DA")] , x = "TestWk" , y = "units" , dots_on = TRUE , colDotgroup = "TestWk" , dots_color = "black" , dots_alpha = 0.4 , dots_size = 0.8 , notch = TRUE , fill = "TestWk.dates" , facet_formula=c("country_code ~ TreatmentWk1") , yscale = "log" , legend="top" , legend.color=TRUE ) } printToPDF(P.EU_pricingtest_units, height=12) { ggBoxplotWithDots( # DT = DT.EU_aggd[transac_type_abbr %in% c("DT", "DA")][] DT = DT.EU_aggd[store_name == "iTunes"] , x = "TestWk" , y = "revenue" , colDotgroup = "TestWk" , notch = TRUE , fill = "TestWk" , facet_formula=c("TreatmentWk1 ~ country_code") ) + scale_y_log10() + coord_flip() # + ylim(0, 3) } DT.EU_aggd[, units.logit := car::logit(units)] ggplot(DT.EU_aggd[transac_type_abbr == "DA"]) + geom_density(aes(x=units.logit)) ggplot(DT.EU_aggd[transac_type_abbr == "DA"]) + geom_density(aes(x=units)) qplot(data=DT.EU_aggd, aes(x=units))