Similar to Who Sets the Agenda: Media or Parliament?: A panel data study to the agenda setting effects on attention to migration and integration - Panel data
Adrian Kay - The Dynamics of Public PolicyDadang Solihin
Similar to Who Sets the Agenda: Media or Parliament?: A panel data study to the agenda setting effects on attention to migration and integration - Panel data (20)
Who Sets the Agenda: Media or Parliament?: A panel data study to the agenda setting effects on attention to migration and integration - Panel data
1. Who Sets the Agenda: Media or Parliament?
A panel data study to the agenda setting effects on
attention to migration and integration
Panel data
Assignment 7
Mark Boukes (markboukes@Hotmail.com)
5616298
1st semester 2010/2011
Dynamic Data Analysis
Lecturer: Dr. R. Vliegenthart
January 12, 2010
Communication Science (Research MSc)
Faculty of Social and Behavioural Sciences
University of Amsterdam
5. Introduction
In this study I aim to investigate the influence that news coverage about issues of migration
and integration has on parliamentary attention to this topic; the agenda setting power of the
media on politics will investigated. However, it is also logical to expect a agenda setting
effect in the contrary direction. Therefore, I will also investigate if parliamentary attention for
issues of migration and integration stimulates the attention of newspapers for such topics.
Therefore, my two research questions are:
• Is parliamentary attention to issues of migration and integration caused by media
attention to this topic?
• Is media attention to issues of migration and integration caused by parliamentary
attention to this topic?
Method
To answer these questions panel data will be used. Because of the time components in the
repeated measures, it is possible to see if an change in the independent variable precedes
changes in the dependent variable. Consequently, we can be more sure that there is a causal
relation and not a third variable that influences both variables at the same time. In order to
investigate whether changes in media attention to issues of migration and integration have an
influence on parliamentary attention, data was gathered for both of these processes. Roggeband
and Vliegenthart (2007) have already done this, and their data was used for this study.
The data for media attention was gathered via a computer-assisted content analysis,
which was conducted using the digital archive of the Web-based version of LexisNexis. They
searched for articles in the five most-read Dutch national newspapers (De Telegraaf,
Algemeen Dagblad de Volkskrant, NRC Handelsblad and Trouw) between 1995 and 2004,
the period in which they were interested. The search engine Parlando was used to obtain data
for parliamentary attention. Parlando contains all documents discussed in and presented to
Parliament and Senate, and allows it thus to create the variable for parliamentary attention to
issues of migration and integration. For both variables a monthly basis was chosen by
Roggeband and Vliegenthart for the period from 1995 to 2004. Furthermore, they split their
variables up into five issues. Those are frames by which media or politicians spoke or wrote
about migration and integration: a multicultural frame, an emancipation frame, a restriction
frame, a victimisation of women frame and an Islam-as-threat frame. A total of 5,376 frames
were found in the sample for parliamentary attention, and a total of 14,972 articles were found
about migration and integration, which contained on average 1.11 frames per article
(Roggeband & Vliegenthart, 2007). Next to the two variables for media and parliamentary
1
6. attention a dummy variable was included in the dataset for the terrorist attacks in New York
on September 11, 2001, to control for a possible increased attention at that moment for the
immigrants, Islam, etc.
To analyse the effects of attention in the media for the five issues of migration and
integration on parliamentary attention to those issues, a multilevel regression analysis is
conducted using Stata 10.1 for the time series of these variables, with issue as a level-2
variable in which observations are nested. Monthly observations are thus cross-classified in
both time and issue. Analyzing this in a multilevel way thus controls for bias caused by
unobserved heterogeneity; in this case recurrent differences in attention for the different
issues, unobserved effects.
Results
In this results section, I specify how the analysis was conducted and discuss the results that
were found. First, a fixed-effects analysis was conducted, which removes cross-sectional
variation; it eliminates the unobserved effect. This fixed-effects regression is a method to
control for omitted variables, when those are constant over time but differ across entities (here
issues) (Stock & Watson, 2003). The fixed effects model gives the same results as conducting
an ordinary least squares regression with dummy variables for the issues. Thereafter, a
random-effects analysis was conducted in which the unobserved effect was subsumed to be a
disturbance term. This analysis is more efficient as the model has less parameters (in this case
four: five dummies minus one, to avoid perfect multicollinearity) (Rabe-Hesketh & Skrondal,
2005). However, two conditions need to be satisfied. First, each observation should be
randomly drawn from a population. The data used for this study uses a sample that constitutes
the whole population of all news articles in the newspapers of interest and all debates and
presentations in Parliament; therefore, it is not a biased sample, and the first condition is
satisfied. The second condition is that unobserved variables are distributed independently
from observed independent variables. This condition will be tested later on with the Hausman
specification test.
The dataset is strongly balanced as there is an observation for every unit (issue) for
every time period. Furthermore, Fisher tests for panel unit root using an augmented Dickey-
Fuller test reject the null hypothesis of the presence of non-stationarity for both the media
attention variable (χ2 = 123.22, p < 0.001) and the parliament attention variable (χ2 = 138.63,
p < 0.001). Thus, it was not necessary to integrate the data.
2
7. Fixed effects analysis
The fixed effects analysis was conducted two times: once with media attention as dependent
variable and once with parliamentary attention as dependent variable, the independent
variable was the lagged value of the variable that was not used as dependent variable (media
or parliament attention), and finally the control variables were the 9/11-dummy, time and time
squared. The last two should control for differences as a consequence of time, either linear or
quadratic. The results of both fixed effects regression models can be found in Table 1.
Table 1. Fixed effects models for either media or parliamentary attention for migration and integration
Parliamentary attention Media attention
Constant 72.433 (35.036)* -847.487 (386.823)*
Media attention(t - 1) 0.015 (0.004)**
Parliamentary attention (t - 1) 2.056 (0.451)**
9/11-dummy -1.878 (0.640)** 47.256 (6.904)**
Month -0.330 (0.150)* 3.763 (1.657)*
Month2 0.000 (.000)* -0.004 (0.002)*
Note. Unstandardized coefficients. Standard errors in parentheses. Month starts in January 1995.
** p < 0.01, * p < 0.05
It seems that parliamentary attention to issues of migration and integration is affected
significantly in a positive way by media attention (F(1, 586) = 17.34, p < 0.001); on average
and holding other variables constant, a one article increase in the number of articles in the five
newspapers, would lead in the next month to 0.015 more discussions in Parliament. To make
it clearer, when the media attention increases with 66 articles in a month; that is about 2
articles a day and 0.4 per newspaper a day, one extra discussion about migration or
integration will take place in the upcoming month. On the other hand, media attention seems
also to be caused by parliamentary attention in a significant and positive way (F(1, 586) =
20.79, p < 0.001); as in one month the attention in Parliament goes up with one discussion or
presentation, the newspapers will on average publish two more articles about migration or
integration in the next month; that is about 0.06 article a day. Thus, when both effects are
compared it seems that the effect of the media on Parliament seems relatively to be stronger.
Remarkable is that the effect of the 9/11 terrorist attacks has a positive effect on media
attention, but a negative effect on parliamentary attention. After the attacks about 47 more
articles are published per month about migration and integration, while the number of debates
in Parliament reduced with almost two per month. A similar result was found for the time
variable. It seems thus that the terrorist attacks stimulated the debate about migration and
3
8. integration in the media, but not in Parliament, and that this debate got more media attention
in the course time, while for politicians it became less important over time.
Random effects analyses
As written above, the analyses are repeated here with a random effects analysis, because this
is a more efficient way (less degrees of freedom are lost). The same variables are used as in
the fixed effects models. The result of the random effects analyses are shown in Table 2.
Table 2. Random effects models for either media or parliamentary attention to migration and integration
Parliamentary attention Media attention
Constant 71.998 (35.098)* -844.873 (387.360)*
Media attention(t - 1) 0.015 (0.004)**
Parliamentary attention (t - 1) 2.018 (0.451)**
9/11-dummy -1.846 (0.641)** 47.194 (6.911)**
Month -0.328 (0.150)* 3.751 (1.659)*
Month2 .000 (.000)* -0.004 (0.002)*
The results of the random effects analyses are almost the same as the ones obtained via the fixed
effects analysis. Media attention to the issues of integration and migration still has a positive
and significant effect on the parliamentary attention in the next month (F(1, 586) = 15.78, p <
0.001) and also the effect of parliamentary attention to issues of migration and integration on
media attention to this topic stays similar (F(1, 586) = 20.05, p < 0.001). Because these effects
are so similar to the ones found in the fixed effects analysis just as the effects of the 9/11
terrorist attacks and time-effects, it is not necessary to specify them here again.
Which analysis to use?
To find out if it is possible to use the estimates of the random effects analyses, the two above
specified conditions need to be specified. It was already explained that the first condition,
observations are randomly drawn from a given population, does not pose any problems. The
second condition is whether the unobserved effect is distributed independently of the
independent variables in the model. To check whether this is true, Hausman specification tests
are conducted. The null hypothesis of both regression models cannot be rejected. This means
that the unobserved heterogeneity is distributed independently of the independent variables in
the model, for the model with parliamentary attention as dependent variable (χ2 = 3.68, p =
0.298) and for the model with media attention as dependent variable (χ2 = 5.77, p = 0.123).
Differences in estimates between the two models are thus not systematic and fixed effects are
for that reason inefficient; random effects estimates will not be subject to unobserved
4
9. heterogeneity bias. Therefore we can use the estimates of the random effects analysis, which
is preferred, because constant characteristics for each unit (issue in this case) are retained in
that regression model contrary to the fixed effects model.
To check if the even more simple OLS regression could be used in stead of the random
effects analysis, the Breusch and Pagan Lagrangian multiplier test was conducted for both
models to check if there are unobserved effects at all, which the random effects analysis will
take into account. For the model with parliamentary attention as dependent variable (χ2 =
725.75, p < 0.001) as well as for the model with media attention as dependent variable (χ2 =
2818.33, p < 0.001) the presence of random effects was found. Random effects analyses seem
thus the right way to estimate our models.
The results of those estimates lead to the two following models1:
Parliamentary attention = 71.998 + 0.015*Media attention(t - 1) ─ 1.846*9/11-dummy ─
0.328*Month + 0.000*Month2
Media attention = -844.873+ 2.018*Parliamentary attention(t - 1) + 47.194*9/11-dummy +
3.751*Month ─ 0.004*Month2
Conclusion
This study has found that increases in the number of articles about issues of migration and
integration in the newspapers De Telegraaf, Algemeen Dagblad de Volkskrant, NRC
Handelsblad and Trouw results in increased parliamentary attention to this topic in the next
month. An agenda setting effect of media on politics was thus found. However, also an effect
in the contrary direction was found: an agenda setting effect of politics on media. When the
politicians in parliament spent more attention to migration and integration, also an increase in
media attention to this topic is expected. These conclusions are based on analyses of panel
data with random effects analyses, so both cross-sectional and time series dimension could be
taken into account.
Reference
Rabe-Hesketh, S. & Skrondal, A. (2005). Multilevel and longitudinal modeling using Stata.
College Station (TX): Stata Press.
Roggeband, C., & Vliegenthart, R. (2007). Divergent framing: The evolution of the public
debate on migration and integration in the Dutch Parliament and media, 1995-2004.
West European Politics, 30(3), 524-548.
Stock, J. W., & Watson, M. W. (2003). Introduction to Econometrics. Boston (MA): Addison
Wesley.
1
The exact value for the coefficients belonging to Month2 are respectively 0.000391 and -0.0041199
5
10. Do File
use H:DDAframes_pooled
findit xtfisher
codebook frame
tsset frame nr, monthly
gen ny=0
replace ny=1 if nr>499
gen n_sq=nr*nr
twoway (tsline media, lcolor(red)) (tsline politics, lcolor(green)
lpattern(dash) lwidth(medthick)) if frame==1
twoway (tsline media, lcolor(red)) (tsline politics, lcolor(green)
lpattern(dash) lwidth(medthick)) if frame==2
twoway (tsline media, lcolor(red)) (tsline politics, lcolor(green)
lpattern(dash) lwidth(medthick)) if frame==3
twoway (tsline media, lcolor(red)) (tsline politics, lcolor(green)
lpattern(dash) lwidth(medthick)) if frame==4
twoway (tsline media, lcolor(red)) (tsline politics, lcolor(green)
lpattern(dash) lwidth(medthick)) if frame==5
xtfisher media
xtfisher politics
xtserial politics media ny nr n_sq
xtserial media politics ny nr n_sq
***Fixed effects***
xtreg politics l.media ny nr n_sq, fe
test l.media
xtreg media l.politics ny nr n_sq, fe
test l.politics
xi: regress politics l.media nr n_sq i.frame
xtreg politics l.media ny nr n_sq, fe
predict politicsfe, e
estimates stor fixed_effects_p
xtreg media l.politics ny nr n_sq, fe
predict mediafe, e
estimates stor fixed_effects_m
***Random effects***
xtreg politics l.media ny nr n_sq, re
test l.media
xtreg media l.politics ny nr n_sq, re
test l.politics
xtreg politics l.media ny nr n_sq, re
predict politicsre, e
estimates stor random_effects_p
xtreg media l.politics ny nr n_sq, re
predict mediare, e
6
11. estimates stor random_effects_m
hausman fixed_effects_p random_effects_p
hausman fixed_effects_m random_effects_m
xtreg politics l.media ny nr n_sq, re
xttest0
xtreg media l.politics ny nr n_sq, re
xttest0
7