Question 3 (35 marks) Recorded music is under threat due to piracy. We want to understand the main drivers of demand for recorded music, either in digital format (e.g., streaming or downloads) or physical format (e.g., Compact Discs). To do this, we would like to develop a linear regression model with as the dependent variable the total weekly revenue from record sales per music artist, aggregated across the different formats (digital and physical). The main independent variables are listed in Table 1 below. We have a panel data set for the German music market observed across 6.5 years, from mid 2010 to the end of 2016, for a total of 337 weeks. The data track the 387 most successful national and international music artists, including Adele, Madonna and Ed Sheeran. TABLE 1: Variables and Operationalisations Variable Definition Source Mean Min. Max. (SD Dependent variables RecordRevenues Record revenues in thousand euros for GIK Entertainment 22.29 0 3,490.49 artist i (1=1....,387) in week t (t=1,...,337) (58,08) Independent variables Piracy, Number of German Internet users (in G/K Brennerstudie 3.64 2.90 4.40 millions) having illegally downloaded (.49) music from the Internet Fame Cumulative number of album chart offiziellecharts.de 4.21 35.81 placements (=chartposition:") from 1995 (Dewan and (6.43) until week t for artist i Ramaprasad 2012) MusicQualityin Average number of Amazon stars across amazon.de 4.26 5.00 all products for artist i in week t (.63) New Releases Dummies for artist i that are 1 in the musicbrainz .02 1.71 week of the new release, and 0 else Wikipedia (.12) Advertisingn Thousands of euros spent on advertising Ebiquity 2.95 2199.74 on artist i in week t (25.34) Airplayi Number of times artist i is played on Nielsen Music 88.08 0 5.777 radio and tv in week Control (267.43] Questions (a) (5 marks) For each of the independent variables, explain the expected effect (positive or negative) on the dependent variable. (b) (5 marks) If you could add three more independent variables to the model, which ones would you pick and why? (c) (10 marks) Suppose we want to estimate a pooled regression model across artists. Please explain (i) how you would have to organise the data; (ii) how many observations the data would have and (ili) whether there is anything you would add tothe pooled regression model, and if so, what? (d) (15 marks) Suppose the model has been estimated, but unfortunately there are issues with statistical validity, predictive validity and face validity. The results tables are shown below, but they are left empty. Please make up (invent) numbers that indicate issues (problems) with all elements of statistical validity, predictive validity and face validity. Copy the tables below into your answer sheet, add made-up coefficients and other statistics and use arrows to explain each of the validity issues in the outputs. Table 2: Model Summary R Square Durbin-Watson Dependent Variable: Record Revenues Table 3: Coefficients Unstandardized Collinearity Coefficients Statistics Model B Sig. VIF (Constant) Piracy Fame MusicQuality NewRelease Advertising Airplay Dependent Variable: Record Revenues Table 4: Model Summary for Estimation Sample (2010-2014) and Validation Sample (2015-2016) R Year 2014 Selected) (Unselected) Dependent Variable: Record Revenues