Consumer Usage of Broadband Internet Services - Adoption, Usage, and Global Impact of Broadband Technologies - page 212

time is less valuable for lower income and
less educated households. This time prefer-
ence could lead lower income, less educated
customers to use more dialup Internet than
broadband, leaving open the possibility that
once a customer has adopted broadband,
the customer's usage might be positively
correlated with income and education.

The frequency of responses is available upon
request.
8 We note at the outset of our analysis that
some of the data represents subjective opin-
ions and impressions of the respondents,
including statements about what they intend
to do. For example, in the next section we
examine respondents' statements about their
intent to change service providers. Survey
respondents' answers to such questions can
be imprecise. Cummings et al. (1995) find
that when survey respondents are asked
whether a product is worth a particular
price, more respondents will say “yes” if
they are told they have no obligation to buy
the product at the stated price than if there is
an obligation to purchase at the price. In the
case of the survey data we use, respondents'
statements concerning their satisfaction, us-
age and uses, and intent to switch providers
might be imprecise. If the errors in their
answers are random, then the effect on our
research is to decrease the confidence we
can place in our statistical results leading us
to understate the validity of our findings. If
the errors are systematic, in other words, if
respondents consistently understate usage,
then the effect on our research is to bias our
results either up or down, depending on the
direction of error. Since we cannot know
whether respondents made errors in their
answers and, if they did, the direction of
those errors, we cannot do better than report
our results with this caveat.
9 Utility also depends on a random disturbance
that has extreme value distribution of type
I, which gives rise to logit models of choice
between alternatives.
10 A multinomial logit is appropriate for any
binary dependent variable y i and a continuous
independent variable x i , Pr ( y i =1)= F (x i ' b)
where, as before, b is a vector of parameters
to be estimated and F is the logistic cumula-
tive distribution function.

3

Pereira and Ribeiro (2006).

4

ITIF 2008.

5

For detailed information on data used, analy-
ses, and results, please see the full paper,
available at www.purc.ufl.edu.

6

Specifically, data are from the following
surveys: 2006 Consumo de Banda Larga
survey (broadband consumption survey,
abbreviated as BCS), 2008 Inquérito ao
Consumo das Comunicações Electrónicas
survey (survey of consumption of electronic
communication services, abbreviated as
ICSCE), and 2008 Índice Europeu da Satis-
fação de Consumidor (European consumer
satisfaction index, abbreviated as ECSI).
The surveys differ and therefore cannot be
combined for empirical analysis (i.e., data
is cross-sectional). Information on the man-
ner in which the surveys were given and the
households selected is available from the
authors upon request.

7

Nomadic users are characterized based on
their responses to 2008 ICSCE survey ques-
tion Q.93, “What type of Internet connection
do you use at home?” Responses were as
follows: modem or ISDN (1), ADSL or other
XDSL access (2), cable (3), mobile phone
or PDA with broadband Internet connection
(4), phone connected to Internet through
narrowband (5), broadband wireless connec-
tions other than mobile phone and PDA (6),
data transmission cards (7), and other or do
not know, which also were coded responses.
By these categories, respondents answering
(6) were considered nomadic users (those
answering 4 and 7 were considered mobile).