After examining the MRP estimates and disaggregation means of preference in every city/county, I look at the stability of public opinion. A simple regression allows us to assess the association between two variables. In this study, they consist of the attitudes toward social welfare across the same administrative units at two time points. Table 1 presents the results for the four waves of MRP estimates predicted by the prior ones. Notice that I predict the 2012 estimates based on the 2010 data for 23 cities/counties, and use the 2012 estimates for the 20 cities/counties to predict the 2013 estimates.
Table 2. Continuity of Public Opinion toward Social Welfare, 2009-2013.
2009 2010 2012 2013
Intercept 106.437*** 60.767*** 0.354 0.478***
(25.530) (2.606) (0.430) (0.021) 2007 Preference −0.635$
(0.334)
2009 Preference −0.108*
(0.045)
2010 Preference −0.001
(0.008)
2012 Preference 0.145*
(0.065)
R2 0.147 0.217 0.001 0.219
Adj. R2 0.106 0.180 -0.047 0.176
Num. obs. 23 23 23 20
Data source: Tsai (2006), Tsai (2009), Tsai (2012).
Notes: ***p < 0.001$, **p < 0.01$, *p < 0.05$, $p < 0.1.
The first column of Table 2 shows the weak correlation between the 2009 and 2010 estimates. The p-value is 0.07. The correlation between the 2009 and 2010 MRP estimates is statistically significant. The coefficient is -0.108 and the standard error is 0.045.
The negative signs of the regression coefficients mean that there has been a decreasing demand for social welfare; support for more social welfare steadily decreases in 23 cities/counties.
Using the 2010 estimates to predict the 2012 estimates, however, gives rise to the null result. Nevertheless, the 2012 estimates successfully predict the 2013 estimates; the coefficient is 0.145 and the standard error is 0.065. The negative coefficients of the 2009 and 2010 models suggest that the general public has been against more social welfare from 2007 through 2010. The positive coefficient in the 2013 model, however, implies that citizens have changed their minds. On the one hand, it may reflect the economic downturn in 2013 when the economic growth rate was less than 2%.
The general public could react to the worsening economic situation by asking for more subsidies. On the other hand, there may be a cycle of attitudes; citizens vacillate in terms of their attitudes every few years. This pattern confirms Wlezien’s theory (1995) that people will turn around their preferences when they feel there has been too much government spending in certain policy fields. Knowing the real cause of the change in public opinion will allow policy-makers to respond to it. The bottom line of this finding is that the social welfare preference in time t-1 has unstable impacts on preferences in time t across the period of time. This data analysis shows mixed results regarding the stability of public opinion in Taiwan.
VI. Validation
How well do we measure citizens’ preferences toward social welfare? Are there any indicators related to them? It is necessary to conduct a validation test before drawing a conclusion. In considering Tsai and Yu’s (2011) finding that social welfare opinion failed to predict government spending, I assume that the “demand side” is more important to the support for social welfare. Because the government has not released the statistics in 2013, I only test four years of the MRP estimates. First of all, I use the percentage of households with low income to predict the estimated public opinion.
Table 3 gives rise to mixed results. Only in 2009 can the independent variable predict the dependent variable, the MRP estimate of social welfare support.
Table 3. Validation of Estimates by Low-income Households
2007 2009 2010 2012
Intercept 76.371*** 57.028*** 54.607*** 29.669***
(1.035) (1.585) (0.393) (2.808) 2007 Low Income 0.002
(0.077)
2009 Low Income 0.231$
(0.120)
2010 Low Income −0.019
(0.030)
2012 Low Income 1.046
(1.302)
R2 0.000 0.150 0.019 0.035
Adjusted R2 -0.048 0.110 -0.028 -0.019
N 23 23 23 20
Notes: ***p < 0.001, **p < 0.01, *p < 0.05, $p < 0.1.
Data source: Tsai (2006), Tsai (2009), Tsai (2012).
Table 4. Validation of Estimates by Household Average Income
2007 2009 2010 2012
Intercept 74.150*** 73.837*** 52.549*** 38.547***
(3.395) (4.374) (1.184) (7.944) 2007 Household Income 0.022
(0.032)
2009 Household Income −0.159***
(0.042)
2010 Household Income 0.020$
(0.011)
2012 Household Income −0.064
(0.071)
R2 0.022 0.407 0.128 0.043
Adjusted R2 -0.025 0.379 0.087 -0.010
N 23 23 23 20
Notes: ***p < 0.001, **p < 0.01, *p < 0.05, $p < 0.1.
Data source: Tsai (2006), Tsai (2009), Tsai (2012).
It is apparent that the percentages of low-income households have a weak association with the MRP estimates, and thus I turn to the average income of households in multiples of NTD10,000 dollars. Table 4 shows that in 2009 and 2010 the average household income predicts the MRP estimates effectively but in different ways. In 2009, the higher the average income of the city/county, the less support for social welfare that there is. In 2010, however, high household income also means support for social welfare.
The results partially confirm the validity of the MRP estimates;
they are still not perfect. Certainly, it is necessary to develop better theories of social welfare before conducting more validation tests.
Moreover, the average household income may not precisely measure the demand for social welfare. At this stage, I would argue that the validity of the MRP estimates needs more checks but their reliability is out of the question.
VII. Conclusion
This paper has presented empirical evidence regarding the stability of aggregate public opinion. By investigating macro-level preferences toward social welfare across three years, I have found that the public opinion in twenty-three or twenty cities/counties is consistent. MRP implemented using Gibbs sampling as a means of Bayesian inference generates more than 15,000 random draws from the sample for each city/county. The median of a posterior distribution is chosen as the predicted probability of supporting social welfare in each city/county. Then the medians are post-stratified over the sub-groups, gender, education, age, and the interaction term of age and education using the census data in 2000 and 2010. The MRP estimates indeed correct the extreme values of some cities/counties with smaller sample sizes.
To summarize, this paper has four points to make:
1. MRP generates estimates that are more centralized than the original data points. Because MRP partially pools observations across cities, the estimates of sub-national public opinion are stronger than those obtained through the disaggregation of survey data.
2. The validity of the MRP estimates is arguable. There is no established theory to explain individual or aggregate support for
social welfare, and therefore we ought to search for better indicators to verify the MRP estimates.
3. The stability of public opinion toward social welfare is not confirmed. Even though the coefficients between the 2007 and 2009, 2009 and 2010, and 2012 and 2013 MRP estimates are significant, the size is small and the direction is either negative or positive.
4. We shall apply this model to other policy issues. As the literature review indicates, the causal relationship between policy and public opinion may not be uniform across the board. Examining policy preferences across different fields only helps us approach the true preferences of the citizens.
The over-arching question is whether the general public leads policy-making or that policy represents preferences. My findings suggest that the general public could be responsible for public policies because public opinion is stable. The question is whether the general public is informed by policies or that they can formulate policies. It is necessary to trace public opinion over longer periods and to correlate it with more policy indicators. With a rigorous estimation method like MRP, we should not refrain from investigating the complicated causal stories of public opinion and government policies, which would only benefit the prospects of democracy.
There are several limitations of this study as I noted earlier in this paper. While this research produces less uncertain estimates of public opinion toward social welfare, using a single policy to argue for the stability of general policy preferences is at best incomplete.
I believe that the class of public policies can be studied in the
same way and am confident that the findings would hold for other policies. Until then, however, the generalization of the stability of opinion among people in Taiwan is limited.
Moreover, this study focuses on public opinion and neglects the aspect of government. Although Tsai and Yu (2011) have investigated the link between policy and preferences, more systematic research is definitely essential for this discipline.
We should also consider other demographic variables, such as occupation, to predict support for government spending. While the Bayesian statistics can reduce the uncertainty associated with the estimates by incorporating a prior distribution, better data may make the inferences more precise. Finally, the hidden internal division between two merging administrative units (e.g., Tainan City and County) may cause the fluctuation in policy preferences to be under-estimated. With a longer period of data, we can probably find the degree to which administrative mergers equalize the policy preferences of city and county.
References
Achen, Christopher H. 1975. “Mass Political Attitudes and the Survey Response.” American Political Science Review 69 (4):
1218-31.
Ansolabehere, Stephen, Jonathan Rodden, and James Snyder, Jr.
2008. “The Strength of Issues: Using Multiple Measures to Gauge Preference Stability, Ideological Constraint, and Issue Voting.” American Political Science Review 102 (2): 215-32.
Converse, Philip E. 1964. “The Nature of Belief Systems in Mass Publics.” In Ideology and Discontent, ed. David E. Apter. New York: Free Press.
Delli Carpini, Michael X. 2005. “An Overview of the State of Citizens' Knowledge about Politics.” In Communicating Politics: Engaging the Public in Democratic Life, eds. Mitchell S. McKinney, Lynda Lee Kaid, and Dianne G. Bystrom and Diana B Carlin. New York, NY: Peter Lang International Academic Publishers.
Delli Carpini, Michael X., and Scott Keeter. 1996. What Americans Know About Politics and Why It Matters. New Haven, CT:
Yale University Press.
Enns, Peter K., and Paul M. Kellstadt. 2008. “Policy Mood and Political Sophistication: Why Everybody Moves Mood.”
British Journal of Political Science 38: 433-54.
Erikson, Robert S., Gerald C. Wright, and John P. McIver. 1993.
Statehouse Democracy: Public Opinion and Policy in the
American States. Cambridge, UK: Cambridge University Press.
Erikson, Robert, Michael B. MacKuen, and James A. Stimson. 2002.
The Macro Polity. Cambridge, UK: Cambridge University Press.
Fell, Dafydd. 2005. Party Politics in Taiwan: Party Change and the Democratic Evolution of Taiwan, 1991-2004. London, UK:
Routledge.
Gelman, Andrew. 2006. “Prior Distributions for Variance Parameters in Hierarchical Models.” Bayesian Analysis 1 (3): 515-33.
Gelman, Andrew, and Jennifer Hill. 2007. Data Analysis Using Regression and Multilevel/Hierarchical Models. Cambridge, UK: Cambridge University Press.
Ghitza, Yair, and Andrew Gelman. 2011. “Deep Interactions with MRP: Presidential Turnout and Voting Patterns Among Small Electoral Subgroups.” http://www.stat.columbia.edu/~gelman/
research/unpublished/deep-inter.pdf (January 14, 2014).
Hsieh, John Fu-sheng, and Emerson M. S. Niou. 1996. “Salient Issues in Taiwan's Electoral Politics.” Electoral Studies 15 (2):
219-35.
Kastellec, Jonathan P., Jeffrey R. Lax, and Justin H. Phillips. 2014.
“Estimating State Public Opinion With Multi-Level Regression and Poststratification using R.” http://www.princeton.
edu/~jkastell/MRP_primer/mrp_primer.pdf (January 10, 2014).
Lax, Jeffrey R., and Justin H. Phillips. 2009. “How Should We Estimate Public Opinion in the States?” American Journal of
Political Science 53 (1): 107-21.
Lupia, Arthur. 1994. “Shortcuts Versus Encyclopedias: Information and Voting Behavior in California Insurance Reform Elections.” American Political Science Review 88 (1): 63-76.
Pacheco, Julianna. 2010. Dynamic Public Opinion and Policy Responsiveness in the American States. Ph. D. diss.
Pennsylvania State University.
Page, Benjamin, and Robert Shapiro. 1992. The Rational Public.
Chicago, IL: University of Chicago Press.
Park, David, Andrew Gelman, and Joseph Bafumi. 2004. “Bayesian Multilevel Estimation with Postratification: State-Level Estimates from National Polls.” Political Analysis 12 (4): 375-85.
Rigger, Shelley. 2001. From Opposition to Power: Taiwan's Democratic Progressive Party. Boulder, CO: L. Rienner Publishers.
Soroka, Stuart, and Christopher Wlezien. 2004. “Opinion Representation and Policy Feedback: Canada in Comparative Perspective.” Canadian Journal of Political Science 37 (3):
531-59.
Soroka, Stuart, and Christopher Wlezien. 2005. “Opinion-Policy Dynamics: Public Preferences and Public Expenditure in the United Kingdom.” British Journal of Political Science 35 (4):
665-89.
Soroka, Stuart, and Christopher Wlezien. 2009. Degrees of Democracy: Politics, Public Opinion, and Policy. Cambridge,
UK: Cambridge University Press.
Tausanovitch, Chris, and Chris Warshaw. 2013. “Measuring Constituent Policy Preferences in Congress, State Legislatures, and Cities.” Journal of Politics 75 (2): 330-42.
Tsai, Chia-hung. 2006. Multi-level Bayesian Analysis of Policy Preference and Political Behavior in Taiwan. National Science Council Research Project. Project number: NSC 95-2414-H-004 -047 –SSS. Taipei: National Science Council.
Tsai, Chia-hung. 2009. Public Opinion and Policy Representation in Taiwan. Measurement and Model. National Science Council Research Project. Project number:NSC 98-2410-H-004-069-MY3. Taipei: National Science Council.
Tsai, Chia-hung. 2012. Policy Orientation, Performance Evaluations and Voting Behavior: Measurement and Model. National Science Council Research Project. Project number: NSC 101-2410-H-004-113-SS3. Taipei: National Science Council.
Tsai, Chia-hung. 2014. “Incumbent Support: Data Analysis Using Bayesian Multilevel Models.” Taiwan Journal of Applied Economics 96: 69-104.
Tsai, Chia-hung, and Eric C.H. Yu. 2011. “Evaluating the Responsiveness of Local Governments in Taiwan: A Case Study of Policies and Public Opinion between 2006 and 2007.”
Taiwan Political Science Review 15 (1): 73-136.
Warshaw, Chris, and Jonathan Rodden. 2012. “How Should We Measure District-Level Public Opinion on Individual Issues?”
Journal of Politics 74 (1): 203-19.
Western, Bruce. 1998. “Causal Heterogeneity in Comparative Research: A Bayesian Hierarchical Modelling Approach.”
American Journal of Political Science 42 (4): 1233-59.
Wlezien, Christopher. 1995. “The Public as Thermostat: Dynamics of Preferences for Spending.” American Journal of Political Science 39 (4): 981-1000.