Impact Of US Trade Policy On 2017 US Presidential Elections

Background and Introduction

The results of the 2017 presidential elections surprised many when the Republican candidate Donald Trump was able to overtake his Democrat counterpart Hillary Clinton to win the position of the Head of State. It was seen that the middle and lower income the main voters base for republicans who were swayed towards the Republicans. One of the key reasons behind this was hypothesized to be due to their lack of faith towards the Democrat party to solve their economic woes which President Trump had promised to address. Trade policies are widely identified to directly affect voter’s decisions especially among the lower income working group who may feel the effect of increased overseas competition. This paper aims to verify from the political survey poll whether trade policies truly played the role of a driving the vote in favor of Trump. The study hypothesized that income and financial worry arising out of the international trade policy of the United States had a significant effect in swaying political support towards Republicans and away from Democrats.

It is hypothesized that the gain in support of the Republicans was an impact of the concerns of the lower income groups and the perception that existing trade policies were working to hamper their livelihood. The study aims to verify this notion and identify any other factors such as faith in Trump and ethnicity also had any impact on the political atmosphere.

The study used a political survey poll data, published by the Pew Research Center in April 2017. It was obtained from the Roper Center. The data of participants who are above the age of 18 were used. The dataset used consisted of a set of 1234 observations. The data variables extracted from the dataset were as follows: “Partysum” which referred to the party towards which the respondent was more inclined from the day of survey and henceforth. The variable had three levels, namely, leaning towards Republican, leaning towards Democrats and no Response or preference. This variable was taken as the dependent variable and the level representing no preference was taken to be the reference level.

The variable is clearly nominal and it is assumed that it follows a multinomial distribution with each level having a particular probability of occurrence which the study tries to capture or determine.

The independent variable considered in the study that are considered from the larger dataset were income level which initially had nine categories, with minimum being the interval less than $10,000 and greatest being greater than $150,000 and an extra category where the respondent refused to disclose. The income level variable was then redefined and re-coded into four groups, with those with less than $30,000 income per year as the lower income groups, those greater than $75,000 as higher income, those in between as middle income and the group which refused to disclose.

Data and Methodology

The ethnicity variable had five levels, namely, “White, non Hispanic”, “Black, non Hispanic”, “Hispanic”, “other” and “Refused to answer”. The variable was re-coded as Black or non-black. This was because the black ethnicity is stated to be among the lowest income groups and also to check whether there may be some racial politics at play or not.

The study then considers a series of opinion based items. First, in variable Q71, it takes into account whether the respondents “In general, do you think that free trade agreements between the U.S. and other countries have been a good thing or a bad thing for the United States?”, the study only took into account those respondents who could give a definitive answer with “bad thing” as response was taken as “1” and “Good thing” as “0”. Secondly, the variable  Q72, takes into account the respondent’s response to the question, “Thinking about the financial situation of you and your family, Do you think free trade agreements have definitely helped, probably helped, probably hurt, or definitely hurt the financial situation of you and your family?”, with again “0” signifying “Bad thing” and “1” as “good thing” . The variable Q81, constitutes of response to the question “Q.81 In your opinion, which generally has more to do with why a person is rich?” or rather it was re-coded as “0” if they felt that hard-work is less important, “1” if they felt only hard work mattered and “2” if they said that both mattered. Finally the variable “Trump” derived from the variable Q1 of the original dataset has two level, namely, “1” if they said they had faith in President Donald Trump and “0” otherwise. The data can be referenced from Pew Research Center (2017).

The dependent variable is categorical or dummy level with three levels and is assumed to have multinomial distribution (k=3). Therefore the multinomial logistic regression was used to fit a regression model to the data and analysis the significance and effects of the independent variables. The analysis was carried out using SPSS using NOMREG (Park, 2015). A dummy variable was first used to run linear regression on the data to check for the VIF values which signify the multi-collinearity. The data was found to have VIF below 1.5 as shown in Results section (Table 1) below. The data had no significant outliers or missing values.

The following table shows the variance inflation factor for independent variables taken together in one model.

Coefficientsa
Model	Collinearity Statistics
Tolerance	VIF
1	ApprovesTrump	.707	1.414
Incomegroup	.988	1.012
Black	.957	1.045
Q.81 In your opinion, which generally has more to do with why a person is rich [READ AND RANDOMIZE]?	.787	1.271
Q72	.682	1.467
Q.71 In general, do you think that free trade agreements between the U.S. and other countries have been a good thing or a bad thing for the United States?	.634	1.577
a. Dependent Variable: Dummy

Results and Analysis

Table 1: Variance Inflation Factors

The variable Q71 or what the respondents think about impact of US trade policy in general was found to have VIF greater than 1.5 and was thus discarded from consideration. Then the remaining variables taken together all showed VIF well within the assumed threshold of 1.5 (Osborne, 2017). These were then used to run the logistic model.

Coefficientsa
Model	Collinearity Statistics
Tolerance	VIF
1	ApprovesTrump	.744	1.344
Incomegroup	.989	1.011
Black	.958	1.043
Q.81 In your opinion, which generally has more to do with why a person is rich [READ AND RANDOMIZE]?	.787	1.271
Q72	.950	1.053
a. Dependent Variable: Dummy

Table 2: VIF of variables in final model

The respondents were 51.5% Democrat leaning, 42.5% as Republican leaning and 5.9% refused to answer or said they had no preference.

Case Processing Summary
	N	Marginal Percentage
Partysum	Dem/lean D	636	51.5%
DK/Ref-no	73	5.9%
Rep/lean R	525	42.5%
Valid	1234	100.0%
Missing	0
Total	1234
Subpopulation	95a

Table 3: Case Summaries of Party leanings (Dependent Variable)

The goodness of fit statistics of the model using the dependent and independent variables as discussed in the data section are given in the following table (George  & Mallery, 2016):

Pseudo R-Square
Cox and Snell	.504
Nagelkerke	.610
McFadden	.402

Table 4: Goodness of Fit of the Model

The Cox-Snell Residual shows that the model fails to capture about 50 percent of the variation. Nagelkerke measure suggests that 61 percent of the variation remains unexplained and McFadden however suggests that about 59.8 percent of the variation is accounted for in the model (Long, 2014). The factors influencing party loyalty most certainly rely on multitudinous factors that have not been taken into account. Nonetheless, the results of the likelihood ration test shows that the variables considered significantly improve the proportion of explained variation as compared to the intercept model (at 0.1 percent level of significance).

Model Fitting Information
Model	Model Fitting Criteria	Likelihood Ratio Tests
-2 Log Likelihood	Chi-Square	df	Sig.
Intercept Only	1162.224
Final	297.384	864.840	10	.000

Table 5: Overall Model Fit

The overall effect of the variables in the model revealed that approval of trump, annual family income, opinion on how trade policy affected their family, opinion on how hard work affects accumulation of wealth, ethnicity have statistically significant effect on their  party preference.( alpha= 0.5 percent)

Likelihood Ratio Tests
Effect	Model Fitting Criteria	Likelihood Ratio Tests
-2 Log Likelihood of Reduced Model	Chi-Square	df	Sig.
Intercept	353.309	55.925	2	.000
ApprovesTrump	789.210	491.826	2	.000
Incomegroup	305.026	7.642	2	.022
Black	321.564	24.180	2	.000
Q81	335.781	38.397	2	.000
Q72	305.637	8.253	2	.016
The chi-square statistic is the difference in -2 log-likelihoods between the final model and a reduced model. The reduced model is formed by omitting an effect from the final model. The null hypothesis is that all parameters of that effect are 0.

Table 6: Likelihood for individual covariate effects

Since the dependent variable has three categorical levels, two sets of model parameter estimates are obtained, one for Democrats and the other for Republicans (Miller, 2017). The No party preference level was taken to be the reference group. The variable ApprovesTrump was statistically significant had the strongest effect on both sets with a negative effect as  its exponential has the value 0.27 suggesting that odds of a person to be a democrat if he or she approve of Donald Trump is 0.3 and for the second model for Republicans odds is 11.3 that one is republican if he or she supports the man. Income was significant for democrats with a higher income group having odds 1.4 to be a democrat however it was not significant for the republican model. Support for trade policy was a significant predictor for the democrat model with a person finding trade policy as helpful having odds of 1.7 to be democrat. This variable was not significant for the republican model however. The respondent being black was found to be statistically significant for the republican model with odds being about 0.3 that he or she is Republican. Another behavioural trait significant is whether they feel hard work can actually get them financial success with a person believing in it positively having 2.3 odds of being republican (Meyers, Gamst & Guarino, 2016).The results are given in the following table.

Parameter Estimates
Partysuma	B	Std. Error	Wald	df	Sig.	Exp(B)	95% Confidence Interval for Exp(B)
Lower Bound	Upper Bound
Dem/lean D	Intercept	1.446	.323	20.039	1	.000
ApprovesTrump	-1.309	.319	16.850	1	.000	.270	.145	.505
Incomegroup	.349	.128	7.429	1	.006	1.418	1.103	1.823
Black	.459	.378	1.474	1	.225	1.583	.754	3.320
Q81	-.333	.262	1.613	1	.204	.717	.429	1.198
Q72	.568	.256	4.916	1	.027	1.765	1.068	2.916
Rep/lean R	Intercept	-.346	.364	.902	1	.342
ApprovesTrump	2.425	.306	62.869	1	.000	11.298	6.205	20.573
Incomegroup	.222	.139	2.548	1	.110	1.248	.951	1.638
Black	-1.210	.485	6.217	1	.013	.298	.115	.772
Q81	.843	.283	8.892	1	.003	2.324	1.335	4.046
Q72	.142	.281	.257	1	.612	1.153	.665	1.998
a. The reference category is: DK/Ref-no.

Table 7: Parameter Estimates of Final Model

Conclusion

The study implies that although the issue with trade policy is not significant for garnering votes for republicans but it surely is significant in depriving the democrats of support. A lower income group is less likely to vote for democrats while a person of black community is less likely to vote for the republicans. Donald Trump stands out as the single most influential factor among the covariates with strong impact on both groups of party support. Conclusively, although a person’s dissatisfaction with trade policy and financial status does not seem to explain loyalty to republicans but it most certainly affects democrats in a negative light which perhaps led to their downfall.

Reference:

George, D., & Mallery, P. (2016). IBM SPSS statistics 23 step by step: A simple guide and reference. Routledge.

Long, J. S. (2014). Regression models for nominal and ordinal outcomes. The SAGE handbook of regression analysis and causal inference, 173-204.

Meyers, L. S., Gamst, G., & Guarino, A. J. (2016). Applied multivariate research: Design and interpretation. Sage publications.

Miller, R. L. (2017). SPSS for social scientists. Macmillan International Higher Education.

Osborne, J. W. (2017). Regression & Linear Modeling.

Park, H. M. (2015). Regression models for ordinal and nominal dependent variables using SAS, Stata, LIMDEP, and SPSS.

Pew Research Center (2017). Pew Research Center: April 2017 Political Survey, Apr, 2017 [dataset]. 31114967, Version 2. Princeton Survey Research Associates International [producer]. Cornell University, Ithaca, NY: Roper Center for Public Opinion Research, Roper Express [distributor], accessed Jul-13-2018.