Dissertation Proposal Presentation

Size: px

Start display at page:

Download "Dissertation Proposal Presentation"

Jacob Parrish
5 years ago
Views:

1 Dissertation Proposal Presentation Economic Growth and Welfare Improvement from High-Speed Internet Theeradej Suabtrirat, PhD in Economics Student. Wednesday, May 6 th, 2015 at pm. Economics Library (Holden Hall 239).

2 Dissertation Committee Dr. Robert P. McComb, Chair of the Committee Dr. Terry von Ende, Committee Member Dr. Masha Rahnamamoghadam, Committee Member Dr. Benaissa Chidmi, Department of Agricultural Economics, Committee Member

3 Three Essays 1. Economics Growth from High-Speed Internet. 2. Internet Adoption of American Households. 3. The Effect of Internet's Price and Speed on Welfare Improvement. For Each Essay: 10 minutes for presentation. 10 minutes for receiving your comments.

4 First Essay: Economics Growth from High-Speed Internet The Internet serves as an electronic trading platform for households and firms. The Internet accounted for about 21 percent of GDP growth over the last five years in developed countries (Du Rausas et.al., 2011). The National Broadband Plan. The plan aims to make high-speed (a.k.a. broadband) internet available everywhere in the United States. Its cost ranges from $23 billion to $350 billion to install internet infrastructures with a download speed of 4 Mbps to 100+ Mbps, respectively.

5 Economics Growth from High-Speed Internet Research Goals Estimate GDP growth from broadband penetration. This dissertation estimates the effect of broadband penetration in the U.S. economy using state-level data. This dissertation informs U.S. policymakers how much economic growth in the last ten years was associated with the past level of broadband penetration.

6 Contribution The previous literature has a large number of studies on OECD or European countries. Too few studies were done at the state-level of the United States. This dissertation improves the results of two research papers. (1) Crandall et.al. (2007) (2) Thompson and Garbacz (2008).

7 Contribution Crandall et.al. (2007) use a cross-sectional regression and data to estimate the effects of broadband penetration on output growth. They find a positive but insignificant relationship. This dissertation s author believes that the output growth is a long-run phenomenon. This dissertation proposes a cross-sectional regression with a longer period of data (from ). The longer data may produce a better result.

8 Contribution Thompson and Garbacz (2008) estimate the impact of broadband penetration by a panel data model using the U.S. state-level data for They find a (surprisingly) significant negative relationship in some of their models. The negative relationship may result from omitted variable bias. This dissertation proposes panel data model with fixed effects to correct this bias. This dissertation finds a significant positive relationship in some models, which is the contribution of this dissertation to the existing literature.

9 Econometric Model This dissertation follows the model of economic growth invented by Barro (1991). The growth of output could be explained by the growth of factors of production. Output = f(technology, Labor, Physical Capital, Human Capital) Their growth is not directly observed. Proxy variables are used to represent their growth.

10 Econometric Model Theory: Output = f(technology, Labor, Human Capital, Physical Capital) Proxy: GSP = f(bbpen, LFP, EduH, GOS, GSP2000, Tax, FEDum) Proxy variables are used to represent the growth in factors of production. GSP2000 and Tax are additional control variables. FEDum are dummies for individual and time fixed effects. GSP BBpen LFP EduH GOS GSP2000 Tax Gross State Product per capita Number of Broadband Lines Broadband penetration = Population Labor force participation rate. Proportion of population whose education is high school or better Gross Operating Surplus per capita. Initial level of Gross State Product per capita. State-local tax burden.

11 Econometric Model Two approaches are proposed. 1. Cross-sectional regression is the traditional approach. It is commonly used since only two data points are needed. 2. Panel data model is cited to be a more successful technique in the previous literature. It allows controlling for unobserved heterogeneity by including individual and time fixed effects.

12 Data Sources Variables Definition Source GSP Gross state product per capita in constant 2000 dollars. (a.k.a. gross domestic product by state per capita) U.S. Bureau of Economic Analysis (BEA) BBpen LFP EduH GOS Tax Broadband penetration. (computed as # of broadband lines with over 200 kbps in at least one direction/ Population) Labor Force Participation. (computed as civilian labor force / civilian non-institutional population) Proportion of state s residents who has high school education or higher. Gross operating surplus by State per capita in constant 2000 dollars. (computed as gross operating surplus by state / Population) State local tax burden (computed as the sum of per capita tax paid to own state and per capita tax paid to other state divided by per capita income.) Federal Communications Commission (FCC) U.S. Bureau of Labor Statistics (BLS) American Community Survey (ACS) data on educational attainment for year Current Population Survey (CPS) data on educational attainment for year U.S. Bureau of Economic Analysis (BEA) Tax Foundation

13 Expected Result * GSP = f(bbpen, LFP, EduH, GOS, GSP2000, Tax, FEDum) 1. Factors of production should have a positive relationship with the output growth. The impact of broadband penetration is of our interest. 2. Initial output level should have a negative relationship with the output growth. According to the growth convergence hypothesis, rich economy has a slower growth than poor one. 3. Tax should have a negative relationship with the output growth. State-local tax burden falls as income rises by the construction of Tax Foundation. 4. FEDum are intercept terms, so their signs are not important.

14 Expected Result GSP=f(Broadband, LFP, EduH, GOS, GSP2000, Tax, FEDum) * Regression diagnostics (such as checking for biasness, consistency, and efficiency) are not done and will be performed after the proposal presentation. Only one model (cross-sectional growth model vs. panel data model) will be chosen for the final defense.

15 Receiving Comments End of the First Essay.

16 Second Essay: Internet Adoption of American Households. Internet adoption refers to a household s decision to purchase an internet connection in home. Purchasing a home internet is the most important step for a household to learn about the benefits of the Internet. About 65% of U.S. households have a home broadband (FCC 2015). vs. The National Broadband Plan targets 100% of U.S. households to have a home broadband.

17 Internet Adoption of American Households. Research Goals Estimate the probability of internet adoption based on household s characteristics. Major barriers are (1) expensive price of an internet connection, (2) irrelevance between the Internet and daily lives, and (3) lack of ownership of computer with internet connectivity. Non-adoption persists in some groups of U.S. population, such as households with low family income, less education, ethnic minorities, senior citizens, rural residents (FCC, 2015). Household s characteristics are very powerful in predicting the probability of internet adoption.

18 Contribution The previous literature frequently creates demographic tabulation. However, the tabulation produces only a simple statistical summary and cannot predict the probability of internet adoption for a given set of household characteristics. The best study was done by ESA (ESA: Economics and Statistic Administration). ESA uses a linear probability model (LPM) to predict the probability of internet adoption. Although an LPM is simple, it suffers from heteroskedastic error term and cannot control the probability to be between 0 and 1. This dissertation suggests using the (non-linear) binary logit model to solve weaknesses of an LPM.

19 Contribution ESA (2011) finds that: 1. Low computer ownership correlates with low adoption rates. 2. The marginal effect of households characteristics are weaker among computer-owning households than that of general households. This dissertation proposes to perform a two-stage logit model to check whether the weaker marginal effect results from sample selection bias or not. The initial estimation was not done but this dissertation s author believes that SAS 9.3 supports such diagnostic.

20 Econometric Model Prob(Adopt=Yes) = f(age, Adults, Metro, Income, Educ, Parent, Cregion, Race) Variable Type: Continuous, Ordinal (use as cont or nominal), Nominal. Adopt Decision to having a home broadband. Yes=1. No=0. Age Age of the respondent (in years) Adults Number of adults in the household (Mostly 1-6 adults). Educ Education level of the respondent. 1=None or < grade 8 (baseline),, 7=Post grad (MA/PhD) Metro Metropolitan level: urban, suburban, rural (baseline). Income Family annual income: <10K (baseline), 10K-20K,, >150K. Parent Parental status. Yes=1. No=0 (baseline). Cregion Census Region: West, South, Northeast, Midwest (baseline) Race White (baseline), Black, Asian, Mixed race, Native American, Hispanic or Other. Note: These variables are based on Home Broadband 2010 Dataset.

21 Data Sources Home Broadband 2010 Dataset This dataset was used to produce initial result of dissertation proposal. It is a telephone survey on about 2,000 households by Pew Internet Project. It asks how household use internet (i.e. read news, banking, social networking). Current Population Survey (CPS) 2011 with Computer and Internet Use Supplement. This dataset will be used for the result of final dissertation defense. The main CPS collects household s demographic data. The supplement asks 50,000+ households regarding how they use computers and internet.

22 Expected Result * * Prob(Adopt=Y) = f(age, Educ, Metro, Income, Adults, Parent, Cregion, Race) Expected Sign: Negative, Positive, Depend on the baseline. Age Educ Metro Income Adults Parent Cregion Race A younger respondent is more likely to adopt. (The Pew survey asks for the youngest adult at home. Teens are likely to use Internet. But their parents may pay the monthly price of the internet connection.) A more educated respondent is more likely to adopt. Urban HHs are more likely to adopt than rural ones. Richer HHs are more likely to adopt than poor ones. Larger HHs are more likely to adopt. (Lower cost per HH member or more chance for at least one member needs the Internet) HHs with kids are more likely to adopt (for child schooling). Northeast HHs are more likely to adopt than MidWest ones. White HHs are more likely to adopt than Black and Hispanic ones. Note: HH stands for households.

23 Receiving Comments End of the Second Essay.

24 Third Essay: The Effect of Internet's Price and Speed on Welfare Improvement. The benefit of broadband penetration is not totally captured by GDP growth. Large portion of the benefit could be in the form of welfare improvement. Welfare improvement is intangible but leads to a better quality of life. For example, people enjoy free knowledge and entertainment on the Internet. Alternatively, they may use a social networking website to connect to their friends and families.

25 The Effect of Internet's Price and Speed on Welfare Improvement. Research Goals Estimate welfare improvement under various scenarios of price and speed. 1. How much does a price increase of internet packages hurt households? How much monetary compensation do they require to restore their original welfare? 2. How much welfare improvement does a household realize from having a home broadband versus not having it? 3. How much welfare improvement would a household earn from receiving a complementary speed upgrade?

26 Contribution The previous literature estimates welfare improvement by using data on household characteristics and monthly price but not connection speed. Failure to include speed may result in omitted variable bias. FCC s survey finds that 50% and 43% of households cite price and speed as the reasons for choosing their current internet package, respectively (FCC 2010). This dissertation brings speed into consideration and suggests that a household is likely to purchase an internet package that offers a faster speed at a better price.

27 Econometric Model Mixed Logit Model Prob(Int=j) = f(price, DSpeed, Age, Metro, Educ, Income, Adults) The j th Internet package. The available Internet packages vary by cities that HHs live in. Below table is the internet packages in San Francisco. Choice ISP Network TechnDownload Upload Price 1_None None None _Dial0 Unreported Dialup _DSL1 AT&T U-Verse DSL _DSL2 AT&T U-Verse DSL _DSL3 AT&T U-Verse DSL _Cable1 Comcast Cable _Cable2 Comcast Cable _Cable3 Comcast Cable _Cable4 Comcast Cable _Fiber1 Astound Fiber _Fiber2 Astound Fiber _Fiber3 Astound Fiber _WL1 LMi.net Fixed Wireless _WL2 LMi.net Fixed Wireless _WL3 LMi.net Fixed Wireless No.1 is outside good. Not purchase an internet connection. No.2-15 are Inside goods. No.2 is Dialup. No.3-15 are High-speed internet.

28 Econometric Model Mixed Logit Model Prob(Int=j) = f(price, DSpeed, Age, Metro, Educ, Income, Adults) Internet package s characteristics: Price Monthy price that HHs pays for their internet package Dspeed Download Speed in Mbps Upload speed is excluded since it is a given feature. HHs cannot choose it. Household s characteristics: (Treated as continuous variables) Age Age of the respondent (in years) Metro Metropolitan level: urban=3, suburban=2, rural=1. Educ Education level of the respondent. 1=None. 7=Post grad. Income Family annual income: <10K, 10K-20K,, >150K. Adults Number of adults in the household Cregion and race are excluded since they cannot be treated as continuous.

29 Data Sources Current Population Survey (CPS) 2011 contains price and HH characteristics Merged Dataset based on Price Cost of Connectivity (COC) 2012 contains price and download speed CPS 2011 and COC 2012 datasets are split and re-merged by city and network technology. Two datasets are one year apart, so price adjustment will be performed. Price serves as the merging key since CPS does not report the name of internet provider. A household paying a higher price is assumed to have purchased a faster speed, and vice versa. COC 2012 and COC 2013 cover nine U.S. cities.

30 Expected Result Prob(Int=j) = f(price, DSpeed, Age, Metro, Educ, Income, Adults) Price should have a negative coefficient since a higher price discourages a purchase. This leads to negative own-price elasticity and positive cross-price elasticity. Dspeed should have a positive coefficient since a faster speed makes the internet surfing more enjoyable. The coefficients of HH characteristics should be similar to those of the second essay.

31 Expected Result The methodology for welfare estimation is adopted from Zhao, Kockelman and Karlstrom (2011). Assuming the linearity of utility function and absence of income effect, Compensated Variation (CV) is estimated as the below equation. CV = 1 α * (ln j e U ij 1 - ln j e U ij 0 ) Where α = -du/dp or marginal utility of money. It is the negative of the price coefficient. Two log-sum term are the denominators of the mixed logit function. CV = ConV$ * (ExpUtil after ExpUtil before ) ConV$ is the special term that converts of utility into dollars. ExpUtil after is expected utility of the after scenario. ExpUtil before is expected utility of the before scenario.

32 Expected Result Compensated Variation (CV) should be positive if the after scenario is less desirable (i.e. internet packages are more expensive) than the before scenario. CV represents the amount of monetary compensation to bring households back to their original utility level (if they need to face an adverse price change). CV can be estimated to answer the research questions previously discussed. 1. How much does a price increase of internet packages hurt households? How much monetary compensation do they require to restore their original welfare? 2. How much welfare improvement does a household realize from having a home broadband versus not having it? 3. How much welfare improvement would a household earn from receiving a complementary speed upgrade?

33 Receiving Comments End of the Third Essay.

Economic Growth from Broadband Penetration, and its Implications on the National Broadband Plan

1 Economic Growth from Broadband Penetration, and its Implications on the National Broadband Plan Theeradej Suabtrirat * Abstract: This paper estimates the effect of broadband penetration on output growth