100 likes | 117 Views
Lock-in and Unobserved Preferences in Server Operating Systems Adoption: A Case of Linux vs. Windows. Discussant: Bin Gu University of Texas at Austin. Strengths. Important research question One phenomenon Current OS adoptions are correlated with past OS adoptions Two explanations
E N D
Lock-in and Unobserved Preferences in Server Operating Systems Adoption: A Case of Linux vs. Windows Discussant: Bin Gu University of Texas at Austin
Strengths • Important research question • One phenomenon • Current OS adoptions are correlated with past OS adoptions • Two explanations • State dependence • Causal connections between past and current adoptions • Lock-in, switching costs, learning costs, network effects, etc. • Unobserved heterogeneity (spurious state dependence) • No causal connections • Different preferences
Strengths • Important research question • One phenomenon • Current OS adoptions are correlated with past OS adoptions • Two explanations • State dependence • Causal connections between past and current adoptions • Lock-in, switching costs, learning costs, network effects, etc. • Unobserved heterogeneity (spurious state dependence) • No causal connections • Different preferences • New empirical approach • Dynamic discrete choice model with predetermined but non-exogenous explanatory variables (AC2003) • Correlation between lagged dependent variable (previous adoption) and unobserved individual effects
Strengths • Important research question • One phenomenon • Current OS adoptions are correlated with past OS adoptions • Two explanations • State dependence • Causal connections between past and current adoptions • Lock-in, switching costs, learning costs, network effects, etc. • Unobserved heterogeneity (spurious state dependence) • No causal connections • Different preferences • New empirical approach • Dynamic discrete choice model with predetermined but non-exogenous explanatory variables (Arellano and Carrasco 2003) • Correlation between lagged dependent variable (previous adoption) and unobserved individual effects • Great dataset • Establishment level data
Strengths • Important research question • One phenomenon • Current OS adoptions are correlated with past OS adoptions • Two explanations • State dependence • Causal connections between past and current adoptions • Lock-in, switching costs, learning costs, network effects, etc. • Unobserved heterogeneity (spurious state dependence) • No causal connections • Different preferences • New empirical approach • Dynamic discrete choice model with predetermined but non-exogenous explanatory variables (Arellano and Carrasco 2003) • Correlation between lagged dependent variable (previous adoption) and unobserved individual effects • Great dataset • Establishment level data • Surprising results • No state dependence
Comments • Differentiation of the two effects • Traditional approach • State dependence: • Utility evaluations changes with past adoptions • Unobserved heterogeneity: • Random or fixed effects to model unobserved difference across firms • AC 2003 • State dependence • Utility evaluations changes with past adoptions • Unobserved heterogeneity • Individual heterogeneity conditional upon past adoptions
Comments • Differentiation of the two effects • Traditional approach • State dependence • Utility evaluations changes with past adoptions • Unobserved heterogeneity • Random or fixed effects to model unobserved difference across firms • AC 2003 • State dependence • Utility evaluations changes with past adoptions • Unobserved heterogeneity • Individual heterogeneity conditional upon past adoptions • Comparison with other models on unobserved heterogeneity and state dependence • Keane (1997) • Elrod (1988) • Jones and Landwehr (1988) • Steckel and Vanhonacker (1988) • McCulloch and Rossi (1996) • Erdem (1996)
Comments • Binary vs. multinomial choice model • Binary choice model at the segment level • Adoption decision is considered for each type of OS • Multinomial choice model at the server level • Estimation of multinomial choice model from aggregate data
Comments • Binary vs. multinomial choice model • Binary choice model at the segment level • Adoption decision is considered for each type of OS • Multinomial choice model at the server level • Estimation of multinomial choice model from aggregate data • Adoption versus testing • Small-scale adoptions of Linux are more like to be trials than actual adoptions. • Testing demonstrates negative state dependence which cancels out positive state dependence from lock-in effects
Comments • Binary vs. multinomial choice model • Binary choice model at the segment level • Adoption decision is considered for each type of OS • Multinomial choice model at the server level • Estimation of multinomial choice model from aggregate data • Adoption versus testing • Small-scale adoptions of Linux are more like to be trials than actual adoptions. • Testing demonstrates negative state dependence which cancels out positive state dependence from lock-in effects • Network effects • Between segments • E.g. Windows PCs + Windows Server • Within segments • E.g. Filing sharing among Windows PCs