Analysis Two Was there a change in the number of units after changing the price? * By Catalog * By Territory How much money did we make before the wholesale price change How much money did we make after the wholesale price change How much money did we make after the experiment How much could we make across all catalog How much could we make across all Frontline ----------------------------------------------- toTest <- c("country_code", "category", "TreatmentWk1", "discretized_tracks", "OverridePerc", "multi_cd") transac <- "DT" transac <- "DA" value.var <- "deltarevenue" value.var <- "deltaunits" groupCols <- c("country_code", "discretized_tracks", "relative_to_default", "OverridePerc", "multi_cd") INFO <- sprintf("Looking at change in %s for %s\n goruping by %s", ifelse(value.var=="deltarevenue", "REVENUE", "UNITS"), ifelse(transac=="DT", "TRACKS", "ALBUMS"), pasteQand(groupCols)) cat(INFO, "\n") ## Add group column DT.EU_aggd.delta[, group := do.call(paste, c(.SD, sep=" - ")), .SDcols = groupCols] # setnames(DT.EU_aggd.deltaunits, gsub(" Experiment", "_Experiment", names(DT.EU_aggd.deltaunits))) if (!haskey(DT.EU_aggd)) stop ("DT.EU_aggd.delta has lost its key") DT.EU_aggd.deltaunits <- dcast.data.table(DT.EU_aggd.delta[store_name == "iTunes" & transac_type_abbr == transac], makeFormula(setdiff(key(DT.EU_aggd.delta), "TimePeriod"), "TimePeriod"), value.var=value.var) ## CONFIRM: Values as expected stopifnot(identical( DT.EU_aggd.deltaunits[, sumn(`Before_Experiment`), keyby=list(category, country_code, TreatmentWk1)] , DT.EU_aggd.delta[store_name == "iTunes" & transac_type_abbr == transac][TimePeriod=="Before_Experiment", sumn(get(value.var)), keyby=list(category, country_code, TreatmentWk1)] )) ## INTERACTION PLOTS if (FALSE) { par(mfrow=c(1, 2)) DT.EU_aggd.deltaunits[, interaction.plot(x.fa=TreatmentWk1, response=During_Experiment, trace.fa=category)] DT.EU_aggd.deltaunits[, interaction.plot(x.fa=category, response=During_Experiment, trace.fa=TreatmentWk1)] par(mfrow=c(1, 2)) DT.EU_aggd.deltaunits[, interaction.plot(x.fa=TreatmentWk1, response=During_Experiment, trace.fa=country_code)] DT.EU_aggd.deltaunits[, interaction.plot(x.fa=country_code, response=During_Experiment, trace.fa=TreatmentWk1)] } ## Country by Country basis --- ignore p values { cat (".......... Comparing Before & During on a country-by-country basis. Ignore p values .......... \n") for (ctry in names(countries_using)) { cat ("\n\n ----------------------- ", ctry , " ------------------------\n") print(anova(lm(`During_Experiment` ~ TreatmentWk1, data=DT.EU_aggd.deltaunits[country_code == ctry]))) print(anova(lm(`During_Experiment` ~ TreatmentWk1 * category, data=DT.EU_aggd.deltaunits[country_code == ctry]))) print(anova(lm(`During_Experiment` ~ TreatmentWk1 * discretized_tracks, data=DT.EU_aggd.deltaunits[country_code == ctry]))) print(anova(lm(makeFormula("During_Experiment", c("TreatmentWk1", setdiff(groupCols, "country_code")), "*"), data=DT.EU_aggd.deltaunits[country_code == ctry]))) } } invisible(DT.EU_aggd.deltaunits[, {cat ("\n\n-----------\n"); print(do.call(paste, c(.BY, sep = " - "))); cat (" --- samples: " , nrow(.SD) ," \n"); if (nrow(.SD) < 4) cat("Not enought data") else print(anova(lm("During_Experiment ~ TreatmentWk1", data=.SD))); TRUE}, by=c(groupCols[c(1, 2, 5)])]) ## Quick overview of different models cat (".......... Comparing Before & During with and without country_code .......... \n") anova(lm(`Before_Experiment` ~ category * TreatmentWk1 , data=DT.EU_aggd.deltaunits)) anova(lm(`Before_Experiment` ~ category * TreatmentWk1 * country_code , data=DT.EU_aggd.deltaunits)) anova(lm(`During_Experiment` ~ category * TreatmentWk1 , data=DT.EU_aggd.deltaunits)) anova(lm(`During_Experiment` ~ category * TreatmentWk1 * country_code , data=DT.EU_aggd.deltaunits)) anova(lm(`During_Experiment` ~ TreatmentWk1 * country_code , data=DT.EU_aggd.deltaunits)) ## With discretized_tracks and OverridePerc anova(lm(`During_Experiment` ~ category * TreatmentWk1 * country_code * discretized_tracks * OverridePerc , data=DT.EU_aggd.deltaunits)) anova(lm(`During_Experiment` ~ TreatmentWk1 * country_code * discretized_tracks * OverridePerc , data=DT.EU_aggd.deltaunits)) anova(lm(`During_Experiment` ~ TreatmentWk1 * country_code * discretized_tracks * relative_to_default * OverridePerc , data=DT.EU_aggd.deltaunits)) # ## MODEL (ORIGINALY) USING # ## Note that the country IS significant. TreatmentWk1 as well. However, category is NOT significant # Model.During__treatment_country_category <- lm(`During_Experiment` ~ TreatmentWk1 * country_code * category, data=DT.EU_aggd.deltaunits) # Anova.During__treatment_country_category <- anova(Model.During__treatment_country_category) # Tukey.During__treatment_country_category <- TukeyHSD(aov(Model.During__treatment_country_category)) # print(names(Tukey.During__treatment_country_category)) # as.data.table(Tukey.During__treatment_country_category)[["TreatmentWk1:country_code"]][from_country_code == to_country_code] # tail(as.data.table(Tukey.During__treatment_country_category), 1)[[1]][from_country_code == to_country_code] [order(upr)] # plot(Tukey.During__treatment_country_category) # ## MODEL WITH TRACKS & OVERRIDES # ## Note that the country IS significant. TreatmentWk1 as well. However, category is NOT significant # Model.During__treatment_country_tracks_overrides <- lm(`During_Experiment` ~ TreatmentWk1 * country_code * discretized_tracks * OverridePerc, data=DT.EU_aggd.deltaunits) # Anova.During__treatment_country_tracks_overrides <- anova(Model.During__treatment_country_tracks_overrides) # Tukey.During__treatment_country_tracks_overrides <- TukeyHSD(aov(Model.During__treatment_country_tracks_overrides)) # print(names(Tukey.During__treatment_country_tracks_overrides)) # as.data.table(Tukey.During__treatment_country_tracks_overrides)[["TreatmentWk1:country_code"]][from_country_code == to_country_code] # tail(as.data.table(Tukey.During__treatment_country_tracks_overrides), 1)[[1]][from_country_code == to_country_code] [!(zero_in_interval)] # plot(Tukey.During__treatment_country_tracks_overrides) ## MODEL WITH OVERRIDE PERC -- no TRACKS ## Note that the country IS significant. TreatmentWk1 as well. However, category is NOT significant Model.During__treatment_country_tracks <- lm(`During_Experiment` ~ TreatmentWk1 * country_code * relative_to_default * OverridePerc * multi_cd, data=DT.EU_aggd.deltaunits) Model.During__treatment_country_tracks <- lm(`During_Experiment` ~ TreatmentWk1 * country_code * discretized_tracks * relative_to_default , data=DT.EU_aggd.deltaunits) Model.During__treatment_country_tracks <- lm(`During_Experiment` ~ TreatmentWk1 * country_code * relative_to_default * OverridePerc , data=DT.EU_aggd.deltaunits) ## USING: Model.During__treatment_country_tracks <- lm(`During_Experiment` ~ TreatmentWk1 * country_code * relative_to_default , data=DT.EU_aggd.deltaunits) Anova.During__treatment_country_tracks <- anova(Model.During__treatment_country_tracks) Tukey.During__treatment_country_tracks <- TukeyHSD(aov(Model.During__treatment_country_tracks)) tmp_DT.ll <- as.data.table(Tukey.During__treatment_country_tracks) cat(INFO, "\n") for (tmp_nm in names(tmp_DT.ll)) { cat (" \n\n----------------- ", tmp_nm, " ---------------\n\n") tmp_DT <- tmp_DT.ll[[tmp_nm]] froms <- setdiff(extract("from_", tmp_DT), "from_TreatmentWk1") if (length(froms)) # print(tmp_DT[Reduce("&", lapply(froms, function(fr) get(fr) == get(gsub("from_", "to_", fr))))] [!(zero_in_interval)] [orderch(names(tmp_DT)[c(3, 5)])] ) print(tmp_DT[Reduce("&", lapply(froms, function(fr) get(fr) == get(gsub("from_", "to_", fr))))] [`p adj` < 0.1] [orderch(names(tmp_DT)[c(3, 5)])] ) # tmp_DT[(`p adj` < .05)] else # print(tmp_DT [!(zero_in_interval)] [orderch(names(tmp_DT)[c(3, 5)])] ) print(tmp_DT [`p adj` < 0.005] [orderch(names(tmp_DT)[c(3, 5)])] ) }; cat("\n", INFO, "\n") &&& RESULTS &&& --------------- 10.99 / 9.99 Stat significant from 8.99 but not from each other GB Different from ES/DE/FR Multi CDs in Germany --------------- &&& RESULTS &&& &&&& LEFT OFF HERE &&&&&&&&& DT.EU_aggd.deltaunits[, pairwise.t.test(x=During_Experiment, g=TreatmentWk1)] DT.EU_aggd.deltaunits[, pairwise.t.test(x=During_Experiment, g=country_code)] &&&&&&&&& # if (FALSE) { # ## REMINDER TO SELF: # ## # ## "During" modeled agains "After" makes no sense. They are NOT INDEPENDENT. # ## By Definition: "During" := "Wk.0" - "Wk.-1" # ## "After" := "Wk.1" - "Wk.0" # ## They both depend on Wk.0 !! # ## # anova(lm(`During_Experiment` ~ `Before_Experiment` * category * TreatmentWk1 * country_code, data=DT.EU_aggd.deltaunits)) # ggplot(DT.EU_aggd.deltaunits, aes_string(x="`Before_Experiment`", y="`During_Experiment`", color="TreatmentWk1")) + geom_point(alpha=0.3) + legendtop() # } ## ----------------------------------------- ## # setnames(DT.EU_aggd.deltaunits, gsub("_Experiment", "\nExperiment", names(DT.EU_aggd.deltaunits))) ## ----------------------------------------- ## notch <- TRUE setkeyIfNot(DT.EU_aggd.delta, "store_name", "transac_type_abbr") P.deltaunits_new <- DT.EU_aggd.delta[transac_type_abbr != "DV", { list(list( ggBoxplotWithDots(DT.EU_aggd.delta[.BY], x="TimePeriod", y="deltaunits", color="TimePeriod", colDotgroup="TimePeriod", facet_formula=". ~ TreatmentWk1", jwidth=0.2, dots_alpha=0.2, ylims=c(-10, 10), title=sprintf("%s, %s", .BY[["store_name"]], .BY[["transac_type_abbr"]]), xlab="", x_angledtext=0.8, notch=notch) + ylim(-15, 15) ))} , keyby=list(store_name, transac_type_abbr) ]$V1 printToPDF(P.deltaunits_new, height=4, width=4, paginate=TRUE, pdfFunc="pdf") ## Color by TimePeriod, x is COUNTRY -- no dots P.by_country <- ggBoxplotWithDots(DT.EU_aggd.delta[(is_iTunesDA)][, TimePeriod := gsub("_", "\n", TimePeriod)], x="country_code", y="deltaunits", color="TimePeriod", colDotgroup="TimePeriod", facet_formula=". ~ TreatmentWk1", jwidth=0.2, dots_alpha=0.2, title=sprintf("%s, %s", .BY[["store_name"]], .BY[["transac_type_abbr"]]), xlab="", x_angledtext=0.8, relativetext=0.7, dots_on = TRUE, violin=FALSE, notch=notch, legend="top", ylims=c(-6, 6)) ## Color by TimePeriod, x is COUNTRY -- no dots P.by_region_group_facet <- P.by_country + facet_grid("discretized_tracks ~ TreatmentWk1") + gg_hline(0, size=0.3, alpha=0.8) printToPDF(P.by_region_group_facet, heigh=40, width=20) ggplot(DT.EU_aggd.delta[(is_iTunesDA )][country_code %in% ctry], alpha=global_alpha) + aes(x=TimePeriod, y=deltaunits, color=country_code, group=upc) + geom_point(alpha=global_alpha) + geom_line(alpha=global_alpha) + facet_grid("category ~ TreatmentWk1") + labs(title=paste0("Album Downloads for ", DT.country[.(ctry)]$country_name), y="Log of Units", x="") + legendtop(notitle=TRUE) + angledtext() + ylim(-20, 20) { ggplot(DT.EU_aggd[(is_iTunesDA & !is.na(TestWk))][country_code %in% ctry], alpha=global_alpha) + aes(x=TestWk.dates, y=units, color=country_code, 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() } ## 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)