SlideShare una empresa de Scribd logo
1 de 13
Media attention in Belgium:

How much influence do citizens and politicians have on
                 news coverage?




                   Pooled time-series analysis




                           Assignment 9




             Mark Boukes (markboukes@Hotmail.com)
                             5616298


                      1st semester 2010/2011
                      Dynamic Data Analysis
                    Lecturer: Dr. R. Vliegenthart
                          February 3, 2010



                                 Communication Science (Research MSc)
                               Faculty of Social and Behavioural Sciences
                                                 University of Amsterdam
Table of contents



INTRODUCTION AND THEORY........................................................................................................................1
METHOD........................................................................................................................................................1
RESULTS........................................................................................................................................................2
    ORDINARY LEAST SQUARES REGRESSION.............................................................................................................................2
    LINEAR REGRESSION MODELS WITH PANEL-CORRECTED STANDARD ERRORS........................................................................................3
    FIXED EFFECT MODELS..................................................................................................................................................5
CONCLUSION.................................................................................................................................................7
REFERENCE....................................................................................................................................................7
APPENDIX A: DO FILE.....................................................................................................................................7
Introduction and theory
In this study, I aim to investigate whether journalists’ attention is as much influenced by
political happenings and previous media coverage, as it is by the actions of ordinary people
undertaken to attract the attention of journalists as well as politicians. For this reason, I will
study the relationship between media attention, parliamentary attention, new legislation,
meetings of the Cabinet and demonstrations by citizens. As it is the duty of journalists to both
inform citizens about what is happening in society and thus in politics but also to function as a
discussion platform for different opinions in society, my hypothesis is:

           Political happenings and demonstrations of citizens have a similar
           influence on news coverage.
This hypothesis will be investigated by means of a secondary data analysis. The same data
will be used as Vliegenthart and Walgrave (2010) used, what makes it possible to conduct a
pooled time-series analysis, which is considered as a robust analysis method. The advantage
of this method is that it captures over-time and between-issue variations, and therefore will
produce more efficient estimates, while it does not exceed assumptions wrongfully.

Method
To investigate the hypothesis, the dataset of Vliegenthart and Walgrave (2010)1 was used,
which contained information for the attention of different actors (media, politics, citizens) to
25 issues. The data was collected on a monthly basis from the beginning of 1993 until the end
of 2000. In total 96 weeks were covered and the attention was divided into 25 issues;
therefore there it was possible to analyse 2400 month-issue combinations. That the data was
collected in such a way that attention was organized in 25 different issues, had the advantage
that it was possible to correct for potential differences in the (error) processes of the different
issues in the estimation techniques that were necessary to investigate the hypothesis. Data was
strongly balanced, as observations were available for every issue for every months and every
actor.
           The attention for the issues was coded in such a way that it was the relative attention
for a certain issue. This means that for every month every actor had a total attention of 1. This
attention will be divided over the 25 issues. For example, an actor can have 0.5 attention for 2
issues in a month, but then cannot have any attention for other issues that month. When one
issue attracts a lot of attention in one month the other issues will thus get less attention. The
analyses need to take care of this contemporaneous correlation.


1
    For more information about how they collected their information, see their article.


                                                                                                      1
Different statistical techniques will be used, to test the robustness of the results and the
necessity of pooling the data. First, an ordinary least squares regression will be conducted,
however, this assumes that the error processes for all issues have the same variance and that
all those error processes are independent of each other, even those of the same issues at
different points in time. This absence of panel heteroscedasticity and contemporaneous
correlation seems very unlikely, but for statistical interest the model will be conducted.
        When analysing time-series cross-section data it is necessary to structure the error
processes in ways to take care of panel heteroscedasticity and contemporaneous correlation
and also unit-level heterogeneity. Therefore, linear regression models with panel-corrected
standard errors are used which perform remarkably well under conditions of panel
heteroscedasticity and contemporaneous correlation (Beck & Katz, 1995). However, panel-
corrected standard errors still have the disadvantage that errors of processes that show unit-
level heterogeneity, are not taken care of. Hence, the less parsimonious fixed effect model
will be conducted to check if and how the results change compared to the model with panel-
corrected standard errors, by allowing different intercepts for the various issues (Wilson &
Bulter, 2007).

Results
Here the results of the various analyses will be presented, starting with the most simple
ordinary least squares via regression that has panel corrected standard errors to fixed effect
models. Before doing the analyses, all variables were tested for the presence of panel unit root
using an augmented Dickey-Fuller test. For all these tests the null hypothesis was rejected (χ2
values were found between 1008.27 and 1733.77), so it was not necessary to integrate the
data.

Ordinary Least Squares regression
Ordinary least squares regression has many assumptions that need to be satisfied before its
results can trusted. One of those is that errors and variances are uncorrelated over the
observations. In time-series cross-section (panel) data this is logically not the case, but to
become aware of the consequences of performing such an analysis on such data, the results of
an ordinary least squares regression will also be presented in this paper. Three regression
analyses will be run that differ in their dynamic specification; the static model with only
variables that were observed in the same month, this model supplemented by distributed lag
independent variables and finally a model that contains the independent variables, the lagged
independent variables and the lagged dependent variable. The results of the analyses can be
found in Table 1.



                                                                                                  2
Table 1. The effects of the (lagged) (in)dependent variables on media attention for an issue
                                           OLS                       DL(1)          ARDL(1,1)
 Media attention t-1                                                              0.782 (0.013)**
 Demonstrations t                    0.088 (0.005)**          0.071 (0.006)**     0.040 (0.004)**
 Demonstrations t-1                                           0.032 (0.006)**    -0.020 (0.004)**
 Parliamentary attention t           0.255 (0.013)**          0.188 (0.013)**     0.072 (0.009)**
 Parliamentary attention t-1                                  0.147 (0.014)**     0.006 (0.009)
 New legislation t                   0.017 (0.007)*          -0.001 (0.007)      -0.003 (0.005)
 New legislation t-1                                          0.009 (0.007)       0.004 (0.004)
 Meeting of the Cabinet t            0.149 (0.015)**          0.082 (0.015)**     0.023 (0.009)*
 Meeting of the Cabinet t-1                                   0.073 (0.015)**     0.005 (0.009)
 Constant                            0.022 (0.001)**          0.019 (0.001)**     0.004 (0.001)**
 Observations (N)                         2400                     2375               2375
 R2                                       0.361                   0.420               0.776
Note. Unstandardized coefficients. Standard errors in parentheses;
* p < 0.05 , ** p < 0.01

As explained above, those estimates are not really trustworthy; therefore, I would only like to
pay attention to the differences between the models. First, the effects of new legislation get
insignificant after adding the lags of the independent variables. Those stay insignificant when
also the autoregressive term (or lagged dependent variable) is added to the model; then also
the lagged effects of parliamentary attention and meetings of the Cabinet get insignificant.
Perhaps more remarkable is that the effect of the lagged independent variable
‘demonstrations’ turns negative. However, I do not pay much attention to interpretation of the
coefficients, as those are not as reliable as the results of estimation methods that take better
care of the regression assumptions. Nevertheless, it is clear that every next model in Table 1
did explain more variance, as R2 increased substantially. Therefore and because it gives a
more complete overview, the variables in the final model, ARDL(1,1), will be used in the next
analyses.

Linear regression models with panel-corrected standard errors
Variance-comparison tests show that, as expected, variance is different for the different
issues: the hypothesis of equality of variances is soundly rejected by all three variance-
comparison tests. This indicates the presence of group-level heteroscedasticity. Consequently,
linear regression models are conducted with panel-corrected standard errors. Those Prais-
Winsten regression analyses can be conducted by different autocorrelation structures: the first
does not take a special autoregressive process in the variance into account, but just a lagged
dependent variable and therefore gives the same results as the last OLS model; one accounts
for first-order autocorrelation AR(1) and assumes that the coefficient of the AR(1) process is


                                                                                                    3
the same to all issues; another specifies that, within panels, there is first-order autocorrelation
and that the coefficient of the AR(1) process is specific to each of the issues. As it seems
plausible that different autoregressive structures can exist for different issues, some might
fade away quicker, because they satisfy the criteria of journalistic institutions less than other
issues (see for example the news criteria specified by Harcup and O'Neill, 2001); therefore,
the last estimation technique seems more helpful. The AR(1) processes seem indeed panel
specific as the rhos (autocorrelation parameters) for the last model vary considerably.
Logically, the proportion of explained variance is also higher for the model that allows the
coefficient of the AR(1) to be specific to each panel. The results of the models can be found in
Table 2.
Table 2. The effects of the (lagged) (in)dependent variables on media attention for an issue
                                     Independent
                                   autocorrelation            General AR(1)          Panel specific AR(1)
                                       structure
Media attention t-1                  0.782     0.020
Demonstrations t                     0.040     0.005          0.043 (0.005)**           0.044  (0.005)**
Demonstrations t-1                  -0.020     0.005          0.006 (0.005)             0.007  (0.005)
Parliamentary attention t            0.072     0.011          0.087 (0.012)**           0.090  (0.012)**
Parliamentary attention t-1          0.006     0.011          0.059 (0.011)**           0.063  (0.011)**
New legislation t                   -0.003     0.005          0.001 (0.005)             0.004  (0.005)
New legislation t-1                  0.004     0.005          0.003 (0.005)             0.006  (0.005)
Meeting of the Cabinet t             0.023     0.012          0.039 (0.012)**           0.045  (0.012)**
Meeting of the Cabinet t-1           0.005     0.012          0.030 (0.012)*            0.039  (0.012)**
Constant                             0.004     0.001          0.031 (0.001)**           0.032  (0.001)**
Observations (N)                          2375                     2375                      2375
R2                                       0.776                    0.123                      0.228
                                                                                         Sample of six:
ρ (autocorrelation parameter)                                       0.695            between 0.484 and 0.788
Note. Unstandardized coefficients. Panel-corrected standard errors in parentheses.
** p < 0.01, * p < 0.05

The first thing to note is that the estimates of the models with the AR(1) structure and the
model with the panel specific AR(1) hardly differ. Only some marginal differences between
the coefficients exist. When we compare these results to the ones found by the ARDL(1,1)
model with the ordinary least squares regression model, four differences appear. First, the
negative effect of the lagged value of demonstrations becomes insignificant. Above, it was
already explained that this negative effect would be a strange finding, so it seems a good sign
that this effect disappears. Furthermore, the coefficients of the lagged values of both
parliamentary attention and meetings of Cabinet become significant. This suggests a stronger
impact of politics on journalists than the findings found by the ordinary least squares



                                                                                                        4
regression technique. Next to this, the proportion of explained variance falls from 0.776 to
0.228.

Fixed effect models
A regression with panel-corrected standard errors can excellently deal with group-level
heteroscedasticity and contemporaneous correlation in a regression analysis. However, it does
not take good enough care of unit-level heterogeneity: systematic differences between values
of different panels (here issues). Indeed it was found that unit-level heterogeneity is present in
the model (F(24, 2341) = 21.30, p < 0.001). Therefore, it was necessary to also conduct fixed
effect model analyses and compare its results with the results found by the linear regression
models with panel-corrected standard errors. The results of three fixed effect models can be
found in Table 3.
Table 3. The effects of the (lagged) (in)dependent variables on media attention for an issue
                                   Static fixed effect                                   Fixed effect model with
                                                             Fixed effect model
                                         model                                             AR(1) disturbance
Media attention t-1                                             0.470 (0.018)**
Demonstrations t                    0.049 (0.004)**             0.039 (0.003)**                0.038 (0.003)**
Demonstrations t-1                                             -0.011 (0.003)**                0.002 (0.003)
Parliamentary attention t           0.074 (0.009)**             0.054 (0.008)**                0.055 (0.008)**
Parliamentary attention t-1                                     0.001 (0.008)                  0.027 (0.008)**
New legislation t                  -0.000 (0.005)              -0.003 (0.004)                 -0.004 (0.004)
New legislation t-1                                             0.002 (0.004)                 -0.003 (0.004)
Meeting of the Cabinet t            0.018 (0.011)               0.009 (0.009)                  0.009 (0.009)
Meeting of the Cabinet t-1                                     -0.008 (0.009)                 -0.002 (0.009)
Constant                            0.035 (0.001)**             0.018 (0.001)**                0.036 (0.001)**
Observations (N)                         2400                        2375                           2350
Overall R2                               0.311                      0.770                          0.347
ψ (variance between issues)              0.033                      0.017                          0.034
θ (variance over time)                   0.021                      0.019                          0.019
ρ (intraclass correlation)               0.709                      0.464                          0.643
Note: Cells contain unstandardized coefficients with standard errors in parentheses. ** p < 0.01
ψ = unexplained variation at country-level, θ = unexplained variation at individual level, ρ = ψ / total variance

All three models take into account that observations are nested within issues. The first is a
standard static fixed effect model that just takes independent variables into account that are
measured at the same time as the dependent variable; the second is a normal fixed effect
model that is similar to the ARDL(1,1) model; the last model includes a AR(1) disturbance
process in which only the previous variance is taken into account, whereas the AR(1)-process
as in the second model continuous to affect media attention in every next step with a
decreasing strength. The estimated coefficients of the three models do not differ a lot, with



                                                                                                                    5
two exceptions. In the normal fixed effects model, the negative effect of lagged
‘demonstrations’ becomes significant again, just as in the ordinary least squares regression
ARDL(1,1) model, while it stays insignificant in the fixed effect model with AR(1)
disturbance. The other difference is that the fixed effect model with AR(1) disturbance finds a
significant effect of the lagged value of parliamentary attention, whereas this was not found
by the normal fixed effect model. When the results are compared to the coefficients found
with the regression with panel-corrected standard errors and panel specific AR(1) process, the
biggest difference is the disappeared significance of the effects of meeting of the Cabinet, but
the same is the presence of a significant lagged effect of parliamentary attention.
       All together the fixed effect model with AR(1) disturbance seems to give rather robust
results as it matches largely with both the regression models with panel-corrected standard
errors and with the normal fixed effect model. The fixed effect models controlled for unit-
level heterogeneity, but it has not the strength to control for group-level heteroscedasticity
(Modified Wald test: χ2 = 3.5*105, p < 0.001) and contemporaneous correlation (Breusch-
Pagan LM test of independence: χ2 = 498.617, p < 0.001), which both were present.
Therefore, it was valuable to compare the results of the fixed effect models with the findings
of the regression with panel corrected standard errors as both have some weaknesses and there
is not an estimation technique that takes all three offences of the OLS assumptions (group-
level heteroscedasticity, contemporaneous correlation and unit-level heterogeneity) into
account.
       The coefficients of this model suggest that both demonstrations and parliamentary
attention have a direct and positive influence on media attention. However, because it is not
possible to see whether those causes precede the effect, it is not totally clear in which
direction the effect runs. Therefore, it is more interesting to see which lagged independent
variables significantly affect media attention. Parliamentary attention is the only lagged
variable that has a significant effect, whereas the other political factors and the citizen factor
do not have a significant lagged effect. It thus seems that politicians have more influence on
media coverage than citizens. A random effect model is not reported in this paper for the
reason that as specified above, it is not expected that unobserved variables are distributed
independently from the observed variables and a Hausman specification test confirmed this
(χ2 = 27.12, p < 0.001).




                                                                                                     6
Conclusion
 Overseeing the strengths and weaknesses of the different estimation techniques and the large
 similarity between the findings of both the fixed effect model and the model with panel
 corrected standard errors, it becomes clear that the power politicians and citizens have on
 journalists is not equal. Though there was found a direct relationship between demonstrations
 and media coverage, a lagged effect could not be proved, which would show that the cause
 preceded the outcome. For political influence, parliamentary attention, such a lagged effect
 was found to be significant. Therefore, the research question needs to be answered negatively;
 political happenings and demonstrations of citizens do not have a similar influence on news
 coverage, politicians have more influence than citizens.

 Reference
 Beck, N., & Katz, J. (1995) What to do (and not to do) with time series cross-section data in
        political science. American Political Science Review, 89(3), 634-647.

 Harcup, T., & O'Neill, D. (2001). What Is news? Galtung and Ruge revisited. Journalism
        Studies, 2(2), 261-280.

 Vliegenthart, R., & Walgrave, S. (2010). When the media matter for politics: Partisan
        moderators of the mass media’s agenda-setting influence on parliament in Belgium.
        Party Politics, 1–22.

 Wilson, S.E., & Butler, D.M. (2007). A lot more to do: The sensitivity of time-series cross-
        section analyses to simple alternative specifications. Political Analysis, 15(2),
        101-123.



Appendix A: Do File
*Assignment9
clear

clear

set memory 250m, permanently
set more off, permanently

use D:DDApanel_belgianas

tsset cissue nr

xtfisher   mediatot_r
xtfisher   legislation_r
xtfisher   parliament_r
xtfisher   min_r
xtfisher   demonstrations_r


                                                                                                 7
*OLS
//normal OLS
regress mediatot_r demonstrations_r parliament_r legislation_r min_r
fitstat

//OLS with distributed lags of the independent variables
regress mediatot_r demonstrations_r l.demonstrations_r parliament_r
l.parliament_r   legislation_r l.legislation_r min_r l.min_r
fitstat

//OLS with distributed lags of the independent variables and lagged
dependent variable = ARDL(1,1)
regress mediatot_r l.mediatot_r demonstrations_r l.demonstrations_r
parliament_r l.parliament_r legislation_r l.legislation_r min_r l.min_r
fitstat
predict double eps, residua
robvar eps, by(cissue)


xtpcse mediatot_r l.mediatot_r demonstrations_r l.demonstrations_r
parliament_r l.parliament_r legislation_r l.legislation_r min_r l.min_r,
correlation(independent)

xtpcse mediatot_r demonstrations_r l.demonstrations_r parliament_r
l.parliament_r legislation_r l.legislation_r min_r l.min_r, correlation(ar1)

xtpcse mediatot_r demonstrations_r l.demonstrations_r parliament_r
l.parliament_r legislation_r l.legislation_r min_r l.min_r,
correlation(psar1)
xttest2

predict y
gen res=mediatot_r-y
xtserial res
corr res l.res l2.res

*fixed effects: static
xtreg mediatot_r demonstrations_r parliament_r legislation_r   min_r, fe


*fixed effects with a AR
xtreg mediatot_r l.mediatot_r demonstrations_r l.demonstrations_r
parliament_r l.parliament_r legislation_r l.legislation_r min_r l.min_r, fe
//contemporenaous correlation: present, yes! negative correlations, if one
party goes up in the polls another will go down.
xttest2

*null Hypothesis: no significant differences between groups. Reject, group
level heteroscedasticity/unit level heterogeneity!
xttest3

* including ar(1) error structure in fe model *
*Or: previous error term influences the next error term, comparable to an AR
term
xtregar mediatot_r demonstrations_r l.demonstrations_r parliament_r
l.parliament_r legislation_r l.legislation_r min_r l.min_r, fe
predict mediafe, e
estimates stor fixed_effects




                                                                              8
xtregar mediatot_r demonstrations_r l.demonstrations_r parliament_r
l.parliament_r legislation_r l.legislation_r min_r l.min_r, re
predict mediare, e
estimates stor random_effects

xtserial mediare
hausman fixed_effects random_effects

findit xtserial




                                                                      9

Más contenido relacionado

Similar a Media attention in Belgium: How much influence do citizens and politicians have on news coverage? - Pooled time-series analysis

Statistical-Process-Control-Analysis-Unraveled_updated210
Statistical-Process-Control-Analysis-Unraveled_updated210Statistical-Process-Control-Analysis-Unraveled_updated210
Statistical-Process-Control-Analysis-Unraveled_updated210
pbaxter
 
Download-manuals-hydrometeorology-data processing-43statisticalanalysiswithr...
 Download-manuals-hydrometeorology-data processing-43statisticalanalysiswithr... Download-manuals-hydrometeorology-data processing-43statisticalanalysiswithr...
Download-manuals-hydrometeorology-data processing-43statisticalanalysiswithr...
hydrologywebsite1
 
Download-manuals-hydrometeorology-data processing-43statisticalanalysiswithr...
 Download-manuals-hydrometeorology-data processing-43statisticalanalysiswithr... Download-manuals-hydrometeorology-data processing-43statisticalanalysiswithr...
Download-manuals-hydrometeorology-data processing-43statisticalanalysiswithr...
hydrologyproject001
 
Impact Analysis V12
Impact Analysis V12Impact Analysis V12
Impact Analysis V12
Bayesia USA
 
Download-manuals-ground water-manual-gw-volume2designmanualsamplingprinciples
 Download-manuals-ground water-manual-gw-volume2designmanualsamplingprinciples Download-manuals-ground water-manual-gw-volume2designmanualsamplingprinciples
Download-manuals-ground water-manual-gw-volume2designmanualsamplingprinciples
hydrologyproject001
 
2014.12.01 - NAEC-Strategic Foresight Workshop_Session 5_Paul Ormerod
2014.12.01 - NAEC-Strategic Foresight Workshop_Session 5_Paul Ormerod2014.12.01 - NAEC-Strategic Foresight Workshop_Session 5_Paul Ormerod
2014.12.01 - NAEC-Strategic Foresight Workshop_Session 5_Paul Ormerod
OECD_NAEC
 

Similar a Media attention in Belgium: How much influence do citizens and politicians have on news coverage? - Pooled time-series analysis (20)

Turn left or right: How the economy affects political preferences and media c...
Turn left or right: How the economy affects political preferences and media c...Turn left or right: How the economy affects political preferences and media c...
Turn left or right: How the economy affects political preferences and media c...
 
Data Analysis Paper
Data Analysis PaperData Analysis Paper
Data Analysis Paper
 
Sustainable development in three newspapers: How does coverage in a particula...
Sustainable development in three newspapers: How does coverage in a particula...Sustainable development in three newspapers: How does coverage in a particula...
Sustainable development in three newspapers: How does coverage in a particula...
 
Pdca prob solving & decision making
Pdca prob solving & decision makingPdca prob solving & decision making
Pdca prob solving & decision making
 
Statistical-Process-Control-Analysis-Unraveled_updated210
Statistical-Process-Control-Analysis-Unraveled_updated210Statistical-Process-Control-Analysis-Unraveled_updated210
Statistical-Process-Control-Analysis-Unraveled_updated210
 
panel data.ppt
panel data.pptpanel data.ppt
panel data.ppt
 
Panel data_25412547859_andbcbgajkje852.ppt
Panel data_25412547859_andbcbgajkje852.pptPanel data_25412547859_andbcbgajkje852.ppt
Panel data_25412547859_andbcbgajkje852.ppt
 
Attention for de Publieke omroep in newspapers: public broadcasting in the n...
Attention for de Publieke omroep in newspapers:  public broadcasting in the n...Attention for de Publieke omroep in newspapers:  public broadcasting in the n...
Attention for de Publieke omroep in newspapers: public broadcasting in the n...
 
Download-manuals-hydrometeorology-data processing-43statisticalanalysiswithr...
 Download-manuals-hydrometeorology-data processing-43statisticalanalysiswithr... Download-manuals-hydrometeorology-data processing-43statisticalanalysiswithr...
Download-manuals-hydrometeorology-data processing-43statisticalanalysiswithr...
 
Download-manuals-hydrometeorology-data processing-43statisticalanalysiswithr...
 Download-manuals-hydrometeorology-data processing-43statisticalanalysiswithr... Download-manuals-hydrometeorology-data processing-43statisticalanalysiswithr...
Download-manuals-hydrometeorology-data processing-43statisticalanalysiswithr...
 
Causality for Policy Assessment and 
Impact Analysis
Causality for Policy Assessment and 
Impact AnalysisCausality for Policy Assessment and 
Impact Analysis
Causality for Policy Assessment and 
Impact Analysis
 
Panel slides
Panel slidesPanel slides
Panel slides
 
A Method for Detection of Outliers in Time Series Data
A Method for Detection of Outliers in Time Series DataA Method for Detection of Outliers in Time Series Data
A Method for Detection of Outliers in Time Series Data
 
Regression models for panel data
Regression models for panel dataRegression models for panel data
Regression models for panel data
 
Answers
AnswersAnswers
Answers
 
Discussion of “Anatomy of sovereign distress: The role of financial sector fr...
Discussion of “Anatomy of sovereign distress: The role of financial sector fr...Discussion of “Anatomy of sovereign distress: The role of financial sector fr...
Discussion of “Anatomy of sovereign distress: The role of financial sector fr...
 
Impact Analysis V12
Impact Analysis V12Impact Analysis V12
Impact Analysis V12
 
Download-manuals-ground water-manual-gw-volume2designmanualsamplingprinciples
 Download-manuals-ground water-manual-gw-volume2designmanualsamplingprinciples Download-manuals-ground water-manual-gw-volume2designmanualsamplingprinciples
Download-manuals-ground water-manual-gw-volume2designmanualsamplingprinciples
 
CASE Network Studies and Analyses 246 - Monetary Policy in Transition: Struct...
CASE Network Studies and Analyses 246 - Monetary Policy in Transition: Struct...CASE Network Studies and Analyses 246 - Monetary Policy in Transition: Struct...
CASE Network Studies and Analyses 246 - Monetary Policy in Transition: Struct...
 
2014.12.01 - NAEC-Strategic Foresight Workshop_Session 5_Paul Ormerod
2014.12.01 - NAEC-Strategic Foresight Workshop_Session 5_Paul Ormerod2014.12.01 - NAEC-Strategic Foresight Workshop_Session 5_Paul Ormerod
2014.12.01 - NAEC-Strategic Foresight Workshop_Session 5_Paul Ormerod
 

Más de Mark Boukes (University of Amsterdam)

Más de Mark Boukes (University of Amsterdam) (16)

Cartoon controversy; why the Danish Mohammed cartoons could be published.
Cartoon controversy; why the Danish Mohammed cartoons could be published. Cartoon controversy; why the Danish Mohammed cartoons could be published.
Cartoon controversy; why the Danish Mohammed cartoons could be published.
 
Turn left or right: Are political preferences in the Netherlands changed by m...
Turn left or right: Are political preferences in the Netherlands changed by m...Turn left or right: Are political preferences in the Netherlands changed by m...
Turn left or right: Are political preferences in the Netherlands changed by m...
 
Public broadcasting; what should it add, what should be its role and what are...
Public broadcasting; what should it add, what should be its role and what are...Public broadcasting; what should it add, what should be its role and what are...
Public broadcasting; what should it add, what should be its role and what are...
 
Journalism and the media: the cartoon controversy: Why were they published?
Journalism and the media: the cartoon controversy: Why were they published?Journalism and the media: the cartoon controversy: Why were they published?
Journalism and the media: the cartoon controversy: Why were they published?
 
Climate change; explaining the differences in reporting
Climate change; explaining the differences in reportingClimate change; explaining the differences in reporting
Climate change; explaining the differences in reporting
 
Explaining political coverage in the Netherlands; Why Dutch voters are less c...
Explaining political coverage in the Netherlands; Why Dutch voters are less c...Explaining political coverage in the Netherlands; Why Dutch voters are less c...
Explaining political coverage in the Netherlands; Why Dutch voters are less c...
 
Journalisten, politici en agendasetting
Journalisten, politici en agendasettingJournalisten, politici en agendasetting
Journalisten, politici en agendasetting
 
Reducing prejudice via mediated contact with immigrants - Proposal for an exp...
Reducing prejudice via mediated contact with immigrants - Proposal for an exp...Reducing prejudice via mediated contact with immigrants - Proposal for an exp...
Reducing prejudice via mediated contact with immigrants - Proposal for an exp...
 
Thematische congruentie en de invloed van advertenties
Thematische congruentie en de invloed van advertentiesThematische congruentie en de invloed van advertenties
Thematische congruentie en de invloed van advertenties
 
Structural equation modelling
Structural equation modellingStructural equation modelling
Structural equation modelling
 
Influence of media source on political interest;
Influence of media source on political interest;Influence of media source on political interest;
Influence of media source on political interest;
 
Interviewing ethnic minorities
Interviewing ethnic minoritiesInterviewing ethnic minorities
Interviewing ethnic minorities
 
Causality
CausalityCausality
Causality
 
Onbewust asocialer
Onbewust asocialerOnbewust asocialer
Onbewust asocialer
 
The emotion-sexual attraction link
The emotion-sexual attraction linkThe emotion-sexual attraction link
The emotion-sexual attraction link
 
Invloed van pacing bij effecten van thematische congruentie op attitude en he...
Invloed van pacing bij effecten van thematische congruentie op attitude en he...Invloed van pacing bij effecten van thematische congruentie op attitude en he...
Invloed van pacing bij effecten van thematische congruentie op attitude en he...
 

Último

{Qatar{^🚀^(+971558539980**}})Abortion Pills for Sale in Dubai. .abu dhabi, sh...
{Qatar{^🚀^(+971558539980**}})Abortion Pills for Sale in Dubai. .abu dhabi, sh...{Qatar{^🚀^(+971558539980**}})Abortion Pills for Sale in Dubai. .abu dhabi, sh...
{Qatar{^🚀^(+971558539980**}})Abortion Pills for Sale in Dubai. .abu dhabi, sh...
hyt3577
 
THE OBSTACLES THAT IMPEDE THE DEVELOPMENT OF BRAZIL IN THE CONTEMPORARY ERA A...
THE OBSTACLES THAT IMPEDE THE DEVELOPMENT OF BRAZIL IN THE CONTEMPORARY ERA A...THE OBSTACLES THAT IMPEDE THE DEVELOPMENT OF BRAZIL IN THE CONTEMPORARY ERA A...
THE OBSTACLES THAT IMPEDE THE DEVELOPMENT OF BRAZIL IN THE CONTEMPORARY ERA A...
Faga1939
 
₹5.5k {Cash Payment} Independent Greater Noida Call Girls In [Delhi INAYA] 🔝|...
₹5.5k {Cash Payment} Independent Greater Noida Call Girls In [Delhi INAYA] 🔝|...₹5.5k {Cash Payment} Independent Greater Noida Call Girls In [Delhi INAYA] 🔝|...
₹5.5k {Cash Payment} Independent Greater Noida Call Girls In [Delhi INAYA] 🔝|...
Diya Sharma
 

Último (20)

05052024_First India Newspaper Jaipur.pdf
05052024_First India Newspaper Jaipur.pdf05052024_First India Newspaper Jaipur.pdf
05052024_First India Newspaper Jaipur.pdf
 
{Qatar{^🚀^(+971558539980**}})Abortion Pills for Sale in Dubai. .abu dhabi, sh...
{Qatar{^🚀^(+971558539980**}})Abortion Pills for Sale in Dubai. .abu dhabi, sh...{Qatar{^🚀^(+971558539980**}})Abortion Pills for Sale in Dubai. .abu dhabi, sh...
{Qatar{^🚀^(+971558539980**}})Abortion Pills for Sale in Dubai. .abu dhabi, sh...
 
BDSM⚡Call Girls in Sector 143 Noida Escorts >༒8448380779 Escort Service
BDSM⚡Call Girls in Sector 143 Noida Escorts >༒8448380779 Escort ServiceBDSM⚡Call Girls in Sector 143 Noida Escorts >༒8448380779 Escort Service
BDSM⚡Call Girls in Sector 143 Noida Escorts >༒8448380779 Escort Service
 
1971 war india pakistan bangladesh liberation.ppt
1971 war india pakistan bangladesh liberation.ppt1971 war india pakistan bangladesh liberation.ppt
1971 war india pakistan bangladesh liberation.ppt
 
Verified Love Spells in Little Rock, AR (310) 882-6330 Get My Ex-Lover Back
Verified Love Spells in Little Rock, AR (310) 882-6330 Get My Ex-Lover BackVerified Love Spells in Little Rock, AR (310) 882-6330 Get My Ex-Lover Back
Verified Love Spells in Little Rock, AR (310) 882-6330 Get My Ex-Lover Back
 
04052024_First India Newspaper Jaipur.pdf
04052024_First India Newspaper Jaipur.pdf04052024_First India Newspaper Jaipur.pdf
04052024_First India Newspaper Jaipur.pdf
 
Enjoy Night⚡Call Girls Rajokri Delhi >༒8448380779 Escort Service
Enjoy Night⚡Call Girls Rajokri Delhi >༒8448380779 Escort ServiceEnjoy Night⚡Call Girls Rajokri Delhi >༒8448380779 Escort Service
Enjoy Night⚡Call Girls Rajokri Delhi >༒8448380779 Escort Service
 
Embed-4.pdf lkdiinlajeklhndklheduhuekjdh
Embed-4.pdf lkdiinlajeklhndklheduhuekjdhEmbed-4.pdf lkdiinlajeklhndklheduhuekjdh
Embed-4.pdf lkdiinlajeklhndklheduhuekjdh
 
THE OBSTACLES THAT IMPEDE THE DEVELOPMENT OF BRAZIL IN THE CONTEMPORARY ERA A...
THE OBSTACLES THAT IMPEDE THE DEVELOPMENT OF BRAZIL IN THE CONTEMPORARY ERA A...THE OBSTACLES THAT IMPEDE THE DEVELOPMENT OF BRAZIL IN THE CONTEMPORARY ERA A...
THE OBSTACLES THAT IMPEDE THE DEVELOPMENT OF BRAZIL IN THE CONTEMPORARY ERA A...
 
Nara Chandrababu Naidu's Visionary Policies For Andhra Pradesh's Development
Nara Chandrababu Naidu's Visionary Policies For Andhra Pradesh's DevelopmentNara Chandrababu Naidu's Visionary Policies For Andhra Pradesh's Development
Nara Chandrababu Naidu's Visionary Policies For Andhra Pradesh's Development
 
BDSM⚡Call Girls in Indirapuram Escorts >༒8448380779 Escort Service
BDSM⚡Call Girls in Indirapuram Escorts >༒8448380779 Escort ServiceBDSM⚡Call Girls in Indirapuram Escorts >༒8448380779 Escort Service
BDSM⚡Call Girls in Indirapuram Escorts >༒8448380779 Escort Service
 
Nurturing Families, Empowering Lives: TDP's Vision for Family Welfare in Andh...
Nurturing Families, Empowering Lives: TDP's Vision for Family Welfare in Andh...Nurturing Families, Empowering Lives: TDP's Vision for Family Welfare in Andh...
Nurturing Families, Empowering Lives: TDP's Vision for Family Welfare in Andh...
 
Enjoy Night ≽ 8448380779 ≼ Call Girls In Gurgaon Sector 47 (Gurgaon)
Enjoy Night ≽ 8448380779 ≼ Call Girls In Gurgaon Sector 47 (Gurgaon)Enjoy Night ≽ 8448380779 ≼ Call Girls In Gurgaon Sector 47 (Gurgaon)
Enjoy Night ≽ 8448380779 ≼ Call Girls In Gurgaon Sector 47 (Gurgaon)
 
Kishan Reddy Report To People (2019-24).pdf
Kishan Reddy Report To People (2019-24).pdfKishan Reddy Report To People (2019-24).pdf
Kishan Reddy Report To People (2019-24).pdf
 
Enjoy Night⚡Call Girls Iffco Chowk Gurgaon >༒8448380779 Escort Service
Enjoy Night⚡Call Girls Iffco Chowk Gurgaon >༒8448380779 Escort ServiceEnjoy Night⚡Call Girls Iffco Chowk Gurgaon >༒8448380779 Escort Service
Enjoy Night⚡Call Girls Iffco Chowk Gurgaon >༒8448380779 Escort Service
 
Gujarat-SEBCs.pdf pfpkoopapriorjfperjreie
Gujarat-SEBCs.pdf pfpkoopapriorjfperjreieGujarat-SEBCs.pdf pfpkoopapriorjfperjreie
Gujarat-SEBCs.pdf pfpkoopapriorjfperjreie
 
₹5.5k {Cash Payment} Independent Greater Noida Call Girls In [Delhi INAYA] 🔝|...
₹5.5k {Cash Payment} Independent Greater Noida Call Girls In [Delhi INAYA] 🔝|...₹5.5k {Cash Payment} Independent Greater Noida Call Girls In [Delhi INAYA] 🔝|...
₹5.5k {Cash Payment} Independent Greater Noida Call Girls In [Delhi INAYA] 🔝|...
 
AI as Research Assistant: Upscaling Content Analysis to Identify Patterns of ...
AI as Research Assistant: Upscaling Content Analysis to Identify Patterns of ...AI as Research Assistant: Upscaling Content Analysis to Identify Patterns of ...
AI as Research Assistant: Upscaling Content Analysis to Identify Patterns of ...
 
Enjoy Night ≽ 8448380779 ≼ Call Girls In Gurgaon Sector 46 (Gurgaon)
Enjoy Night ≽ 8448380779 ≼ Call Girls In Gurgaon Sector 46 (Gurgaon)Enjoy Night ≽ 8448380779 ≼ Call Girls In Gurgaon Sector 46 (Gurgaon)
Enjoy Night ≽ 8448380779 ≼ Call Girls In Gurgaon Sector 46 (Gurgaon)
 
Enjoy Night ≽ 8448380779 ≼ Call Girls In Palam Vihar (Gurgaon)
Enjoy Night ≽ 8448380779 ≼ Call Girls In Palam Vihar (Gurgaon)Enjoy Night ≽ 8448380779 ≼ Call Girls In Palam Vihar (Gurgaon)
Enjoy Night ≽ 8448380779 ≼ Call Girls In Palam Vihar (Gurgaon)
 

Media attention in Belgium: How much influence do citizens and politicians have on news coverage? - Pooled time-series analysis

  • 1. Media attention in Belgium: How much influence do citizens and politicians have on news coverage? Pooled time-series analysis Assignment 9 Mark Boukes (markboukes@Hotmail.com) 5616298 1st semester 2010/2011 Dynamic Data Analysis Lecturer: Dr. R. Vliegenthart February 3, 2010 Communication Science (Research MSc) Faculty of Social and Behavioural Sciences University of Amsterdam
  • 2.
  • 3. Table of contents INTRODUCTION AND THEORY........................................................................................................................1 METHOD........................................................................................................................................................1 RESULTS........................................................................................................................................................2 ORDINARY LEAST SQUARES REGRESSION.............................................................................................................................2 LINEAR REGRESSION MODELS WITH PANEL-CORRECTED STANDARD ERRORS........................................................................................3 FIXED EFFECT MODELS..................................................................................................................................................5 CONCLUSION.................................................................................................................................................7 REFERENCE....................................................................................................................................................7 APPENDIX A: DO FILE.....................................................................................................................................7
  • 4.
  • 5. Introduction and theory In this study, I aim to investigate whether journalists’ attention is as much influenced by political happenings and previous media coverage, as it is by the actions of ordinary people undertaken to attract the attention of journalists as well as politicians. For this reason, I will study the relationship between media attention, parliamentary attention, new legislation, meetings of the Cabinet and demonstrations by citizens. As it is the duty of journalists to both inform citizens about what is happening in society and thus in politics but also to function as a discussion platform for different opinions in society, my hypothesis is: Political happenings and demonstrations of citizens have a similar influence on news coverage. This hypothesis will be investigated by means of a secondary data analysis. The same data will be used as Vliegenthart and Walgrave (2010) used, what makes it possible to conduct a pooled time-series analysis, which is considered as a robust analysis method. The advantage of this method is that it captures over-time and between-issue variations, and therefore will produce more efficient estimates, while it does not exceed assumptions wrongfully. Method To investigate the hypothesis, the dataset of Vliegenthart and Walgrave (2010)1 was used, which contained information for the attention of different actors (media, politics, citizens) to 25 issues. The data was collected on a monthly basis from the beginning of 1993 until the end of 2000. In total 96 weeks were covered and the attention was divided into 25 issues; therefore there it was possible to analyse 2400 month-issue combinations. That the data was collected in such a way that attention was organized in 25 different issues, had the advantage that it was possible to correct for potential differences in the (error) processes of the different issues in the estimation techniques that were necessary to investigate the hypothesis. Data was strongly balanced, as observations were available for every issue for every months and every actor. The attention for the issues was coded in such a way that it was the relative attention for a certain issue. This means that for every month every actor had a total attention of 1. This attention will be divided over the 25 issues. For example, an actor can have 0.5 attention for 2 issues in a month, but then cannot have any attention for other issues that month. When one issue attracts a lot of attention in one month the other issues will thus get less attention. The analyses need to take care of this contemporaneous correlation. 1 For more information about how they collected their information, see their article. 1
  • 6. Different statistical techniques will be used, to test the robustness of the results and the necessity of pooling the data. First, an ordinary least squares regression will be conducted, however, this assumes that the error processes for all issues have the same variance and that all those error processes are independent of each other, even those of the same issues at different points in time. This absence of panel heteroscedasticity and contemporaneous correlation seems very unlikely, but for statistical interest the model will be conducted. When analysing time-series cross-section data it is necessary to structure the error processes in ways to take care of panel heteroscedasticity and contemporaneous correlation and also unit-level heterogeneity. Therefore, linear regression models with panel-corrected standard errors are used which perform remarkably well under conditions of panel heteroscedasticity and contemporaneous correlation (Beck & Katz, 1995). However, panel- corrected standard errors still have the disadvantage that errors of processes that show unit- level heterogeneity, are not taken care of. Hence, the less parsimonious fixed effect model will be conducted to check if and how the results change compared to the model with panel- corrected standard errors, by allowing different intercepts for the various issues (Wilson & Bulter, 2007). Results Here the results of the various analyses will be presented, starting with the most simple ordinary least squares via regression that has panel corrected standard errors to fixed effect models. Before doing the analyses, all variables were tested for the presence of panel unit root using an augmented Dickey-Fuller test. For all these tests the null hypothesis was rejected (χ2 values were found between 1008.27 and 1733.77), so it was not necessary to integrate the data. Ordinary Least Squares regression Ordinary least squares regression has many assumptions that need to be satisfied before its results can trusted. One of those is that errors and variances are uncorrelated over the observations. In time-series cross-section (panel) data this is logically not the case, but to become aware of the consequences of performing such an analysis on such data, the results of an ordinary least squares regression will also be presented in this paper. Three regression analyses will be run that differ in their dynamic specification; the static model with only variables that were observed in the same month, this model supplemented by distributed lag independent variables and finally a model that contains the independent variables, the lagged independent variables and the lagged dependent variable. The results of the analyses can be found in Table 1. 2
  • 7. Table 1. The effects of the (lagged) (in)dependent variables on media attention for an issue OLS DL(1) ARDL(1,1) Media attention t-1 0.782 (0.013)** Demonstrations t 0.088 (0.005)** 0.071 (0.006)** 0.040 (0.004)** Demonstrations t-1 0.032 (0.006)** -0.020 (0.004)** Parliamentary attention t 0.255 (0.013)** 0.188 (0.013)** 0.072 (0.009)** Parliamentary attention t-1 0.147 (0.014)** 0.006 (0.009) New legislation t 0.017 (0.007)* -0.001 (0.007) -0.003 (0.005) New legislation t-1 0.009 (0.007) 0.004 (0.004) Meeting of the Cabinet t 0.149 (0.015)** 0.082 (0.015)** 0.023 (0.009)* Meeting of the Cabinet t-1 0.073 (0.015)** 0.005 (0.009) Constant 0.022 (0.001)** 0.019 (0.001)** 0.004 (0.001)** Observations (N) 2400 2375 2375 R2 0.361 0.420 0.776 Note. Unstandardized coefficients. Standard errors in parentheses; * p < 0.05 , ** p < 0.01 As explained above, those estimates are not really trustworthy; therefore, I would only like to pay attention to the differences between the models. First, the effects of new legislation get insignificant after adding the lags of the independent variables. Those stay insignificant when also the autoregressive term (or lagged dependent variable) is added to the model; then also the lagged effects of parliamentary attention and meetings of the Cabinet get insignificant. Perhaps more remarkable is that the effect of the lagged independent variable ‘demonstrations’ turns negative. However, I do not pay much attention to interpretation of the coefficients, as those are not as reliable as the results of estimation methods that take better care of the regression assumptions. Nevertheless, it is clear that every next model in Table 1 did explain more variance, as R2 increased substantially. Therefore and because it gives a more complete overview, the variables in the final model, ARDL(1,1), will be used in the next analyses. Linear regression models with panel-corrected standard errors Variance-comparison tests show that, as expected, variance is different for the different issues: the hypothesis of equality of variances is soundly rejected by all three variance- comparison tests. This indicates the presence of group-level heteroscedasticity. Consequently, linear regression models are conducted with panel-corrected standard errors. Those Prais- Winsten regression analyses can be conducted by different autocorrelation structures: the first does not take a special autoregressive process in the variance into account, but just a lagged dependent variable and therefore gives the same results as the last OLS model; one accounts for first-order autocorrelation AR(1) and assumes that the coefficient of the AR(1) process is 3
  • 8. the same to all issues; another specifies that, within panels, there is first-order autocorrelation and that the coefficient of the AR(1) process is specific to each of the issues. As it seems plausible that different autoregressive structures can exist for different issues, some might fade away quicker, because they satisfy the criteria of journalistic institutions less than other issues (see for example the news criteria specified by Harcup and O'Neill, 2001); therefore, the last estimation technique seems more helpful. The AR(1) processes seem indeed panel specific as the rhos (autocorrelation parameters) for the last model vary considerably. Logically, the proportion of explained variance is also higher for the model that allows the coefficient of the AR(1) to be specific to each panel. The results of the models can be found in Table 2. Table 2. The effects of the (lagged) (in)dependent variables on media attention for an issue Independent autocorrelation General AR(1) Panel specific AR(1) structure Media attention t-1 0.782 0.020 Demonstrations t 0.040 0.005 0.043 (0.005)** 0.044 (0.005)** Demonstrations t-1 -0.020 0.005 0.006 (0.005) 0.007 (0.005) Parliamentary attention t 0.072 0.011 0.087 (0.012)** 0.090 (0.012)** Parliamentary attention t-1 0.006 0.011 0.059 (0.011)** 0.063 (0.011)** New legislation t -0.003 0.005 0.001 (0.005) 0.004 (0.005) New legislation t-1 0.004 0.005 0.003 (0.005) 0.006 (0.005) Meeting of the Cabinet t 0.023 0.012 0.039 (0.012)** 0.045 (0.012)** Meeting of the Cabinet t-1 0.005 0.012 0.030 (0.012)* 0.039 (0.012)** Constant 0.004 0.001 0.031 (0.001)** 0.032 (0.001)** Observations (N) 2375 2375 2375 R2 0.776 0.123 0.228 Sample of six: ρ (autocorrelation parameter) 0.695 between 0.484 and 0.788 Note. Unstandardized coefficients. Panel-corrected standard errors in parentheses. ** p < 0.01, * p < 0.05 The first thing to note is that the estimates of the models with the AR(1) structure and the model with the panel specific AR(1) hardly differ. Only some marginal differences between the coefficients exist. When we compare these results to the ones found by the ARDL(1,1) model with the ordinary least squares regression model, four differences appear. First, the negative effect of the lagged value of demonstrations becomes insignificant. Above, it was already explained that this negative effect would be a strange finding, so it seems a good sign that this effect disappears. Furthermore, the coefficients of the lagged values of both parliamentary attention and meetings of Cabinet become significant. This suggests a stronger impact of politics on journalists than the findings found by the ordinary least squares 4
  • 9. regression technique. Next to this, the proportion of explained variance falls from 0.776 to 0.228. Fixed effect models A regression with panel-corrected standard errors can excellently deal with group-level heteroscedasticity and contemporaneous correlation in a regression analysis. However, it does not take good enough care of unit-level heterogeneity: systematic differences between values of different panels (here issues). Indeed it was found that unit-level heterogeneity is present in the model (F(24, 2341) = 21.30, p < 0.001). Therefore, it was necessary to also conduct fixed effect model analyses and compare its results with the results found by the linear regression models with panel-corrected standard errors. The results of three fixed effect models can be found in Table 3. Table 3. The effects of the (lagged) (in)dependent variables on media attention for an issue Static fixed effect Fixed effect model with Fixed effect model model AR(1) disturbance Media attention t-1 0.470 (0.018)** Demonstrations t 0.049 (0.004)** 0.039 (0.003)** 0.038 (0.003)** Demonstrations t-1 -0.011 (0.003)** 0.002 (0.003) Parliamentary attention t 0.074 (0.009)** 0.054 (0.008)** 0.055 (0.008)** Parliamentary attention t-1 0.001 (0.008) 0.027 (0.008)** New legislation t -0.000 (0.005) -0.003 (0.004) -0.004 (0.004) New legislation t-1 0.002 (0.004) -0.003 (0.004) Meeting of the Cabinet t 0.018 (0.011) 0.009 (0.009) 0.009 (0.009) Meeting of the Cabinet t-1 -0.008 (0.009) -0.002 (0.009) Constant 0.035 (0.001)** 0.018 (0.001)** 0.036 (0.001)** Observations (N) 2400 2375 2350 Overall R2 0.311 0.770 0.347 ψ (variance between issues) 0.033 0.017 0.034 θ (variance over time) 0.021 0.019 0.019 ρ (intraclass correlation) 0.709 0.464 0.643 Note: Cells contain unstandardized coefficients with standard errors in parentheses. ** p < 0.01 ψ = unexplained variation at country-level, θ = unexplained variation at individual level, ρ = ψ / total variance All three models take into account that observations are nested within issues. The first is a standard static fixed effect model that just takes independent variables into account that are measured at the same time as the dependent variable; the second is a normal fixed effect model that is similar to the ARDL(1,1) model; the last model includes a AR(1) disturbance process in which only the previous variance is taken into account, whereas the AR(1)-process as in the second model continuous to affect media attention in every next step with a decreasing strength. The estimated coefficients of the three models do not differ a lot, with 5
  • 10. two exceptions. In the normal fixed effects model, the negative effect of lagged ‘demonstrations’ becomes significant again, just as in the ordinary least squares regression ARDL(1,1) model, while it stays insignificant in the fixed effect model with AR(1) disturbance. The other difference is that the fixed effect model with AR(1) disturbance finds a significant effect of the lagged value of parliamentary attention, whereas this was not found by the normal fixed effect model. When the results are compared to the coefficients found with the regression with panel-corrected standard errors and panel specific AR(1) process, the biggest difference is the disappeared significance of the effects of meeting of the Cabinet, but the same is the presence of a significant lagged effect of parliamentary attention. All together the fixed effect model with AR(1) disturbance seems to give rather robust results as it matches largely with both the regression models with panel-corrected standard errors and with the normal fixed effect model. The fixed effect models controlled for unit- level heterogeneity, but it has not the strength to control for group-level heteroscedasticity (Modified Wald test: χ2 = 3.5*105, p < 0.001) and contemporaneous correlation (Breusch- Pagan LM test of independence: χ2 = 498.617, p < 0.001), which both were present. Therefore, it was valuable to compare the results of the fixed effect models with the findings of the regression with panel corrected standard errors as both have some weaknesses and there is not an estimation technique that takes all three offences of the OLS assumptions (group- level heteroscedasticity, contemporaneous correlation and unit-level heterogeneity) into account. The coefficients of this model suggest that both demonstrations and parliamentary attention have a direct and positive influence on media attention. However, because it is not possible to see whether those causes precede the effect, it is not totally clear in which direction the effect runs. Therefore, it is more interesting to see which lagged independent variables significantly affect media attention. Parliamentary attention is the only lagged variable that has a significant effect, whereas the other political factors and the citizen factor do not have a significant lagged effect. It thus seems that politicians have more influence on media coverage than citizens. A random effect model is not reported in this paper for the reason that as specified above, it is not expected that unobserved variables are distributed independently from the observed variables and a Hausman specification test confirmed this (χ2 = 27.12, p < 0.001). 6
  • 11. Conclusion Overseeing the strengths and weaknesses of the different estimation techniques and the large similarity between the findings of both the fixed effect model and the model with panel corrected standard errors, it becomes clear that the power politicians and citizens have on journalists is not equal. Though there was found a direct relationship between demonstrations and media coverage, a lagged effect could not be proved, which would show that the cause preceded the outcome. For political influence, parliamentary attention, such a lagged effect was found to be significant. Therefore, the research question needs to be answered negatively; political happenings and demonstrations of citizens do not have a similar influence on news coverage, politicians have more influence than citizens. Reference Beck, N., & Katz, J. (1995) What to do (and not to do) with time series cross-section data in political science. American Political Science Review, 89(3), 634-647. Harcup, T., & O'Neill, D. (2001). What Is news? Galtung and Ruge revisited. Journalism Studies, 2(2), 261-280. Vliegenthart, R., & Walgrave, S. (2010). When the media matter for politics: Partisan moderators of the mass media’s agenda-setting influence on parliament in Belgium. Party Politics, 1–22. Wilson, S.E., & Butler, D.M. (2007). A lot more to do: The sensitivity of time-series cross- section analyses to simple alternative specifications. Political Analysis, 15(2), 101-123. Appendix A: Do File *Assignment9 clear clear set memory 250m, permanently set more off, permanently use D:DDApanel_belgianas tsset cissue nr xtfisher mediatot_r xtfisher legislation_r xtfisher parliament_r xtfisher min_r xtfisher demonstrations_r 7
  • 12. *OLS //normal OLS regress mediatot_r demonstrations_r parliament_r legislation_r min_r fitstat //OLS with distributed lags of the independent variables regress mediatot_r demonstrations_r l.demonstrations_r parliament_r l.parliament_r legislation_r l.legislation_r min_r l.min_r fitstat //OLS with distributed lags of the independent variables and lagged dependent variable = ARDL(1,1) regress mediatot_r l.mediatot_r demonstrations_r l.demonstrations_r parliament_r l.parliament_r legislation_r l.legislation_r min_r l.min_r fitstat predict double eps, residua robvar eps, by(cissue) xtpcse mediatot_r l.mediatot_r demonstrations_r l.demonstrations_r parliament_r l.parliament_r legislation_r l.legislation_r min_r l.min_r, correlation(independent) xtpcse mediatot_r demonstrations_r l.demonstrations_r parliament_r l.parliament_r legislation_r l.legislation_r min_r l.min_r, correlation(ar1) xtpcse mediatot_r demonstrations_r l.demonstrations_r parliament_r l.parliament_r legislation_r l.legislation_r min_r l.min_r, correlation(psar1) xttest2 predict y gen res=mediatot_r-y xtserial res corr res l.res l2.res *fixed effects: static xtreg mediatot_r demonstrations_r parliament_r legislation_r min_r, fe *fixed effects with a AR xtreg mediatot_r l.mediatot_r demonstrations_r l.demonstrations_r parliament_r l.parliament_r legislation_r l.legislation_r min_r l.min_r, fe //contemporenaous correlation: present, yes! negative correlations, if one party goes up in the polls another will go down. xttest2 *null Hypothesis: no significant differences between groups. Reject, group level heteroscedasticity/unit level heterogeneity! xttest3 * including ar(1) error structure in fe model * *Or: previous error term influences the next error term, comparable to an AR term xtregar mediatot_r demonstrations_r l.demonstrations_r parliament_r l.parliament_r legislation_r l.legislation_r min_r l.min_r, fe predict mediafe, e estimates stor fixed_effects 8
  • 13. xtregar mediatot_r demonstrations_r l.demonstrations_r parliament_r l.parliament_r legislation_r l.legislation_r min_r l.min_r, re predict mediare, e estimates stor random_effects xtserial mediare hausman fixed_effects random_effects findit xtserial 9