Multi-Criteria Evaluation of Broadband Internet Access in Poland

Transcription

1 International Workshop on Multiple Criteria Decision Making Multi-Criteria Evaluation of Broadband Internet Access in Poland M. Zawisza 1, B. Kamiński 1, W. Jakuczun 2, A.Gładysz 3 1 Institute of Econometrics, Warsaw School of Economics, 2 WLOG Solutions, 3 Office of Electronic Communications

2 Agenda 1. Motivation 2. Two-step procedure a) Data Envelopment Analysis b) Supervised learning 3. Data description 4. Results a) Data Envelopment Analysis b) Supervised learning 5. Conclusions

3 Motivation Telecommunication regulators around the world state two following research questions: 1. How to measure and compare the advancement of Internet deployment between local markets? 2. How to support (if it is possible?) selected regions in order to imporove their advancement of Internet deployment? Internet deployment advancement across communities

4 Two-step procedure Data Envelopment Analysis (DEA) Data: communities (gminas) from both 2010 and 2011 resulting in decision units characterized by multidimensional attributes Output: single DEA efficiency indices for each community Supervised learning: Explaining DEA efficiency index with the help of variables, which are of partial control by telecom regulator E.g. regression analysis, GLMs, GAMs, ANNs, CARTs

5 DEA step Three objectives, i.e. DEA outputs: #Households with at least one Internet provider (availability) #Households subscribed to Internet (adoption) #Households with at least two providers (competition) Control variables, i.e. DEA inputs: Community size Number of households in a community Input-oriented DEA with increasing returns to scale Implementation: GNU R, Benchmarking package

6 Supervised learning step The explained variable: DEA efficiency index Explaining variables: Computerization level in community schools Local government expenditures on telecommunication Average income in a community Herfindahl-Hirschman Index in a community Aged and educational structure of community population Techniques: regression tree and random forrests Implementation: GNU R, party & randomforest

7 Data sources of communities in 2010 and 2011 Local Markets Dataset of Polish Office of Electronic Communications (UKE) Local Data Bank of Central Statistical Office of Poland (GUS) GUS data covering average income in communities National Census of Population and Housing 2002 provided by GUS Financial budgets of local goverment provided by Agency of Public Information (BIP) of the Ministry of Finance Republic of Poland

8 Results of DEA step Level: low & increasing Diversity: low & decreasing Level: high & stable Diversity: high & decreasing

9 Results of DEA step (cont d) Shift to the right Two-mudal function Shift to the center

10 Results of DEA step (cont d) Level Change Rural Urbanrural Urban

11 Results of supervised learning step What does matter? Community (1, 11) HighEduc (2, 12) WorkAge (5) SchoolComp (15)

12 Results of supervised learning step (cont d)

13 Conclusions We propose a two-step procedure for the Internet performance evaluation of local markets The conducted analysis reveals: Poor performance of rural communities; Improvement of rural communities over ; Decrease in the inequality of Internet deployment. Effective regulation may foster the advancement by the increase of Internet awareness through: Higher level of education in urban communities Higher level of computerization in rural communities