Cognitive ability and party affiliation: The role of the formative years of political socialization

INTELL-01183; No of Pages 7 Intelligence xxx (2017) xxx–xxx

Contents lists available at ScienceDirect

Intelligence

Cognitive ability and party affiliation: The role of the formative years of political socialization☆ Yoav Ganzach Faculty of Management, Tel Aviv University, Israel

a r t i c l e

i n f o

Article history: Received 27 August 2016 Received in revised form 23 December 2016 Accepted 9 January 2017 Available online xxxx

a b s t r a c t We study the effect of time on the relationship between intelligence and party affiliation in the United States. Our results indicate that time affects this relationship, and that this effect is due to the formative years in which political preferences were developed rather than the time in which the survey was conducted. For people who were born in the 20th century, the later their formative years, the more positive the relationship between intelligence and Democratic, as opposed to Republican, affiliation. The current results shed light on recent conflicting findings about the relationship between intelligence and party affiliation in the US, and suggest that the effect of intelligence on party affiliation changes with time. © 2017 Elsevier Inc. All rights reserved.

1. Introduction Studies on the relationship between intelligence and political preferences have focused primarily on the relationship between intelligence and liberal attitudes. Most studies in this area suggest that intelligence is positively correlated with liberal attitudes (Deary, Batty, & Gale, 2008a; Deary, Batty, & Gale, 2008b; Heaven, Ciarrochi, & Leeson, 2011; Hodson & Busseri, 2012; Kanazawa, 2010; Pesta & McDaniel, 2014; Pesta, McDaniel, & Bertsch, 2010; Schoon, Cheng, Gale, Batty, & Deary, 2010; Stankov, 2009). However, because of the complexity of the connection between political attitudes and political behavior, there is growing interest in recent years regarding the relationship between intelligence and another type of political preference – party affiliation, which is more closely related to political behavior. Most of this research was conducted on American samples (but see Deary et al., 2008b and Karadja, Mollerstrom, & Seim, 2014 for studies involving English and Swedish samples, respectively) in which party affiliation is most conveniently defined on a scale ranging from strong affiliation with Republicans to strong identification with Democrats. For simplicity we label this variable Democratic Affiliation (DA). Because liberal attitudes are considered to be associated with DA, the most natural hypothesis is that the relationship between intelligence and DA is positive. Yet Carl (2014a, 2014b) found that this relationship is negative, while Ganzach (2016) found “No important differences” between Democrats and Republicans. (Ganzach, 2016, explains the negative relationship found by Carl as resulting from a lack of control for socio-economic status and racial identity). But even ☆ Financial support was provided by the Coller Institute of Venture and by the Vice President Fund at Tel Aviv University. E-mail address: [email protected].

Ganzach's (2016) findings of few differences leave us with the question of why – given that intelligence is positively correlated with liberal beliefs – are Democrats not more intelligent than Republicans when important background characteristics are controlled for? Our answer is that the intuition that liberal attitudes are associated with Democratic Affiliation is based on recent experience. Although in recent years liberal attitudes are strongly associated with DA, this association was weaker in earlier times (Abramowitz & Saunders, 1998; Levendusky, 2009; Miller & Schofield, 2008. And in particular, see Meisenberg's, 2015, data below).1 Both Carl's (2014a, 2014b) and Ganzach's (2016) conclusions were based on the GSS surveys conducted between 1972 and 2012, years in which the relationship between social and economic attitudes and party affiliation in the American electorate changed considerably (Layman & Carsey, 2002). Thus, the relationship between intelligence and DA in Carl's (2014a, 2014b) and Ganzach's (2016) studies represent aggregate relationships, collapsed over many survey years, which may overlook time-dependent effects underlying the formation of party affiliation, particularly the dependence of the effect of intelligence on time. In the current paper we suggest that in addition to studying the main effect of intelligence on party affiliation, it is also important to examine the interaction between time and intelligence, as this interaction may be an important factor in explaining party affiliation. In particular, if the association between liberal attitudes and DA were weaker in earlier times, we should expect 1 Another answer to the question of why democrats are not more intelligent than republicans was suggested by Carl (2014a), who argued that intelligence is positively associated with both socially liberal beliefs and economically rightist beliefs (i.e., classical liberal beliefs), and that “higher intelligence among classically liberal Republicans compensates for lower intelligence among socially conservative Republicans” (p. 142). However, this answer cannot explain the temporal pattern of the effect of intelligence on DA that is described in the current paper.

http://dx.doi.org/10.1016/j.intell.2017.01.003 0160-2896/© 2017 Elsevier Inc. All rights reserved.

Please cite this article as: Ganzach, Y., Cognitive ability and party affiliation: The role of the formative years of political socialization, Intelligence (2017), http://dx.doi.org/10.1016/j.intell.2017.01.003

2

Y. Ganzach / Intelligence xxx (2017) xxx–xxx

that the association between intelligence and DA would also be weaker at these times. We distinguish between two types of hypotheses regarding the time × intelligence interaction. The first, which we label the period hypothesis, suggests that the effect of intelligence on DA is more positive in later than in earlier periods (i.e., survey years). For example, it suggests that surveys conducted in the early 21st century will show a more positive relationship between intelligence and DA than surveys conducted in the 1970s. If liberal attitudes entice DA, if intelligent people are more liberal, and if the association between liberal attitudes and DA is stronger in later than in earlier survey years, then the association between intelligence and DA should be stronger in later survey years. The second hypothesis, which we label the cohort hypothesis, is based on the idea that young adulthood is a critical period in which political preferences are formed (Alwin, Cohen, & Newcomb, 1991; Peterson, 1983. See Hess, 1973, for a review of the review of critical periods in other domains). It suggests that the effect of intelligence on party affiliation is more positive for people whose formative young adulthood years occurred in later times than on people for whom these years occurred in earlier times. For example, it suggests that the effect of intelligence on DA is stronger for people who became young adults in the 1960s than for people who became young adults in the 1920s. The logic behind this hypothesis is the same as the logic behind the period hypothesis except that the relevant time in which the association between liberal attitudes and DA, and therefore the association between intelligence and DA, is formed, is the time of young adulthood. Meisenberg (2015) took a first step in examining the interaction between time and intelligence. However, Meisenberg's analysis was based on simple correlations between intelligence and political preferences for various periods between 1972 and 2012, and therefore he examined only the interaction between period and intelligence, and in fact confounded this interaction with the interaction between cohort and intelligence. In addition, this correlational analysis does not allow for control of possible confounds, particularly control for socioeconomic status, and does not provide a reliable statistical test for the time × intelligence interaction. Nevertheless, Meisenberg's results are very informative for our subject. In Table 1 I reproduce Meisenberg's (2015) central findings in order to highlight the patterns that are the starting point for the current research. First, the data in this table suggests that the relationship between liberal attitudes and DA became more positive during the years. This positive relationship was rather weak in the 1970s and considerably strengthened in later years (although this trend was weaker

Table 1 Correlations of liberal attitudes with Democratic Affiliation (Lib-DA), intelligence score with liberalism (IQ-Lib), and intelligence score with Democratic Affiliation (IQ-DA). Group 1

Period 2

Lib-DA 3

IQ-Lib 4

IQ-DA 5

White male

1974–1981 1982–1991 1992–2001 2002–2012 1974–1981 1982–1991 1992–2001 2002–2012 1974–1981 1982–1991 1992–2001 2002–2012 1974–1981 1982–1991 1992–2001 2002–2012

0.234⁎⁎⁎ 0.315⁎⁎⁎ 0.437⁎⁎⁎ 0.547⁎⁎⁎ 0.168⁎⁎⁎ 0.285⁎⁎⁎ 0.404⁎⁎⁎ 0.543⁎⁎⁎

0.018 0.010 0.009 0.089⁎⁎⁎ 0.068⁎⁎ 0.086⁎⁎⁎ 0.070⁎⁎⁎ 0.081⁎⁎⁎

−0.135⁎⁎⁎ −0.090⁎⁎⁎ −0.025 −0.025 −0.072⁎⁎ −0.051⁎⁎

0.008 084 140 169⁎⁎⁎

0.138 0.179⁎⁎⁎ 0.036 0.152⁎⁎⁎

−0.066 0.049 0.078 0.153⁎⁎⁎

−0.080 0.091 0.074 0.054

White female

Black male

Black female

⁎ p b 0.05. ⁎⁎ p b 0.01. ⁎⁎⁎ p b 0.001.

−0.004 0.030 0.045 −0.039 0.032 0.035 0.017 0.104⁎⁎ 0.060 0.084

among Blacks than among Whites). This trend is consistent with a general interaction between time and intelligence in which the effect of intelligence on DA increases with time. Second, there is also evidence in Meisenberg's data for a weak trend among Whites (but not among Blacks) for the relationship between intelligence and DA to become more positive during the years. But even these two temporal trends in Meisenberg's (2015) data are mute regarding the question of whether the interaction between time and intelligence is the result of an interaction between period and intelligence or with an interaction between cohort and intelligence. 2. Study 1 2.1. Method 2.1.1. Data Data were taken from the 1972–2012 waves of the General Social Survey (GSS). The GSS collects data on demographic characteristics and attitudes of US residents. The survey is conducted face-to-face with an in person interview of a randomly selected sample of noninstitutionalized adults (18+). The survey has been conducted every year from 1972 to 1994 (except in 1979, 1981 and 1992), and every other year since 1994. The survey takes about 90 min to administer. Thus, as of 2012, 28 national samples with 57,061 respondents and 5417 variables had been collected. Participants were, on average, 45.7.8 (18–89 age range, SD. 17.5) years old. 2.1.2. Measures 2.1.2.1. Intelligence. The GSS measures the verbal intelligence of its respondents by a ten-item multiple-choice measure of vocabulary knowledge called Wordsum. Adding up the number of correct answers yields a total test score. For the case of presentation, raw scores are converted to the commonly used IQ scale with a mean of 100 and standard deviation of 15. Due to the high correlation between verbal intelligence and general intelligence this measure is often used as an indicator of intelligence in GSS research (e.g., Hauser & Huang, 1997; Kanazawa, 2004). 2.1.2.2. Time. In view of our hypotheses, we used two measures of time. Year of Survey (YOS) is the calendar year in which the survey was conducted. Year of Birth (YOB) is the calendar year in which the participant was born. This variable represents, and is directly related, to the participants' formative years of young adulthood (YOB = YOS − Age, where age is the age at the survey year2). For convenience the actual YOS and YOB were divided by 100. 2.1.2.3. Party affiliation. Our measure for party affiliation was based on the question, “Do you think of yourself as a Republican, Democrat, Independent, or what?” This question has eight response categories: “strong Democrat”, “not strong Democrat”, “Independent, near Democrat”, “Independent”, “Independent, near Republican”, “not strong Republican”, “strong Republican” and “other”. We assigned values ranging from 1 to 7 to these responses, the higher the value, the stronger the Democratic Affiliation. We label this variable Democratic Affiliation since the higher the value, the stronger the affiliation with the Democratic party and the weaker the affiliation with the Republican party. 2.1.2.4. Control variables. Education was measured by the number of years of full-time education completed. Income was the log transformed family income in 1986 dollars. Other controls were sex (coded as 1 for 2 Note that YOB, YOS and age are perfect linear functions of each other and therefore their effects cannot be estimated simultaneously. The choice of which effect to estimate is theory-dependent. We further consider this issue in the discussion section.

Please cite this article as: Ganzach, Y., Cognitive ability and party affiliation: The role of the formative years of political socialization, Intelligence (2017), http://dx.doi.org/10.1016/j.intell.2017.01.003

Y. Ganzach / Intelligence xxx (2017) xxx–xxx

3

Table 2 Descriptive statistics – Study 1.

1. Sex 2. Black 3. White 4. Income 5. Education 6. Intelligence 7. YOS 8. YOB 9. DA

Mean

STD

1

2

3

4

5

6

7

8

1.56 0.14 0.81 9.95 12.75 100.00 1992 1946 4.33

0.50 0.35 0.39 1.01 3.18 15.00 12 21 1.99

– 0.05 −0.03 −0.12 −0.04 0.02 0.00 −0.03 0.06

– −0.84 −0.19 −0.11 −0.21 0.02 0.05 0.28

– 0.19 0.10 0.25 −0.10 −0.14 −0.26

– 0.38 0.30 0.00 0.06 −0.13

– 0.48 0.21 0.29 −0.11

– 0.02 −0.03 −0.10

– 0.53 −0.05

– −0.04

Note: n varies between 26,916 and 56,860, depending on missing values.

males and 2 for females), and race (Black, Hispanic, and Caucasian as a comparison group). 2.2. Results and discussion Table 2 presents means, standard deviations and inter-correlations of the study variables. Of particular interest is the negative correlation between DA and intelligence, replicating Carl's result of negative association between the two when no controls are exerted. The correlation between DA and both YOS and YOB is negative, consistent with a decline in affiliation with the Democratic party over time. The correlation between YOS and YOB is high, but as we will see later on, not too high to prevent separating the effects of the two on DA. We begin with a rough comparison of the period × intelligence interaction to the cohort × intelligence interaction using Meisenberg's (2015) approach of examining the correlation between intelligence and DA over relatively long time intervals. The solid line in Fig. 1 depicts the Intelligence-DA correlation for Meisenberg's four YOS time-intervals (see Table 1 and Fig. 1 for specification of the intervals. To decrease noise we calculated these correlations over the entire sample rather than by race/sex). The dashed line depicts these correlations for four time-intervals of YOB that were chosen to include about the same number of participants (see Fig. 1 for specification). The data of this figure suggest that although the intelligence-DA correlation became more positive (less negative) both for YOS and for YOB, this pattern is more pronounced for YOB. This suggests that the cohort hypothesis provides a better description of the data than the period hypothesis. Due to the high correlation of YOS and YOB, there is a considerable overlap between the YOS and YOB time-intervals, and the correlational analysis provides only a crude comparison of the effect of YOS and YOB on the intelligence-DA correlation. Therefore we proceed with a more refined regression analysis. Table 3 presents the results of least square regression models that examine the interactions between time and intelligence on DA.3 Model 1 is a basic model that controls only for demographic variables. Model 2 adds income and education as controls, and Model 3 adds the interactions between time and income and the interactions between time and education as controls.4 In all these models the interaction between YOB and intelligence was highly significant (p b 0.0001), but the interaction between YOS and intelligence was not significant (note also that adding the controls had a very small effect 3 We examined the robustness of our least square regressions using non-parametric regressions. The results were very similar to the results of the least square regressions. The results of ordered-logit regressions are presented in the Appendix 1. 4 The corresponding main effects model showed significant main effects for sex (females were more democratic than males), Black (blacks were more democratic), White (whites were less democratic), YOS (in later surveys people were democratic) and YOB (later cohorts were less democratic). The effect of intelligence was significantly negative only in Model 1, when socio-economic variables were not controlled for, but not significant when these variables were controlled for. This difference is in line with Ganzach (2016) argument that when appropriate controls are exerted, there is no important difference in intelligence between Democrats and Republicans in the GSS data.

on the size of the interaction). Thus the cohort hypothesis is supported, but the period hypothesis is not.5 Fig. 2 plots the interaction between YOB and intelligence (based on Model 2) keeping the other variables constant at their means. It depicts the typical effects of intelligence on DA for older participants whose formative years for development of political preferences were in the 1920′ s (one standard deviation below the average; YOB = 1925) and younger participants whose formative years were in the 1960s (one standard deviation below the average YOB = 1966). For the older participants this relationship is negative, while for the younger participants this relationship is positive. Thus, the overall non-significant relationship between intelligence and DA in the GSS (Ganzach, 2016) may very well be the result of a positive relationship among younger participants and a negative relationship among older participants. Interestingly enough, in Model 3 the interactions between education and YOS and the interaction between income and YOS were significant in the opposite direction. The former interaction was positive, suggesting the effect of education on DA is more positive in later survey years than in earlier survey years. The latter interaction was negative, suggesting that the effect of income on DA is more negative in later survey years than in earlier survey years. The positive interaction between education and YOS is consistent with the idea that the effect of education on DA became more positive in later surveys, when ideology became more important in determining party identity. The negative interaction between income and YOS is consistent with the idea that the effect of income on DA became more negative in later surveys, when income differences in the American society became more important in determining party identity.

3. Study 2 In this study we examine the cohort hypothesis by comparing the relationship between intelligence and DA of two cohorts, an early cohort and a later cohort, that were interviewed in the same year (2008). The formative years of the early cohort were in the late 1970s and early 1980s (mean YOB = 1960, STD = 2.5) whereas the formative years of the later cohort were in the first decade of the 2000s (mean YOB = 1986, STD = 6.4). The cohort hypothesis suggests that this relationship will be more positive among the later cohort than among the earlier cohort.

5 Meisenberg's (2015) results suggest that the effect of time on the intelligence-DA relationship is more positive for Whites than for Blacks (see Column 5 in Table 1). To examine this difference between Whites and Blacks, we estimated Model 3 separately for Whites and Blacks (see Appendix 2). The results showed a significant (p b 0.0001) positive YOB × intelligence interactions for both groups, suggesting that for both the intelligenceDA relationship is more positive for later than for earlier YOBs. The difference between Whites and Blacks in Meisenberg is however consistent with the negative YOS × intelligence interaction for Blacks and a tendency for a positive YOS × intelligence interaction for Whites.

Please cite this article as: Ganzach, Y., Cognitive ability and party affiliation: The role of the formative years of political socialization, Intelligence (2017), http://dx.doi.org/10.1016/j.intell.2017.01.003

4

Y. Ganzach / Intelligence xxx (2017) xxx–xxx

Fig. 2. Democratic Affiliation as a function of low and high Year of Birth (YOB). Low and high YOB are one standard deviation below and above the mean, respectively and they correspond to 1921 and 1967, respectively.

Fig. 1. The Intelligence-DA correlation as a function of period. The YOS (Year of Survey) periods are 1974–1981 (labeled as 1 on the abscissa), 1982–1991 (labeled as 2), 1992– 2001 (labeled as 3) and 2001–2012 (labeled as 4), respectively. The number of observations are 4142, 8142, 7063 and 5021, respectively. The YOB (Year of Birth) periods are b1931 (labeled as 1), 1932–1948 (labeled as 2) 1949–1961 (labeled as 3) and greater than 1962 (labeled as 4). The number of observations are 5960, 6359, 7324 and 6608, respectively.

3.1. Method 3.1.1. Data 3.1.1.1. The early cohort. The data were taken from the 1979 cohort of the National Longitudinal Survey of Youth (NLSY79), conducted with a probability sample of 12,686 Americans (with an oversampling of AfroAmericans, Hispanics and economically disadvantaged whites) born between 1957 and 1964. Natural sample attrition was about 10% a year. The participants were first interviewed in 1979. Until 1994 they were interviewed annually, and from then on they were interviewed every two years. The current study used the results of an intelligence test that was administered to the participants in 1979, and information about their party affiliation derived from the 2008 survey. 3.1.1.2. The later cohort. The data were taken from the NLSY79-Children survey (the NLSYC). The NLSYC is based on the earlier NLSY79 described above. The NLSYC sample frame consists of all children born to female NLSY79 respondents. The number of children who were interviewed

Table 3 Democratic Affiliation and the time × intelligence interactions – Study 1. Model 1

Intercept Sex Black White Intelligence YOS YOB Income Education Income × YOS Income × YOB Education × YOS Education × YOB Intelligence × YOS Intelligence × YOB n R2

Model 2

Model 3

b

stderr

b

stderr

b

stderr

99.21 0.195 0.903 −0.692 −0.728 −0.332 −4.509

14.30 0.024 0.063 0.057 0.141 0.819 0.441

116.4 0.186 0.826 −0.714 −0.896 −1.078 −4.577 −0.107 −0.023

15.2 0.025 0.066 0.060 0.150 0.874 0.480 0.014 0.005

45.49 −0.553 0.183 11.002 −0.713 −5.589 3.463 −5.817 0.829 −0.675⁎⁎ 0.121 0.252⁎⁎

−0.004 0.042⁎⁎

0.008 0.004

0.003 0.043⁎⁎ 23,852 0.098

0.009 0.005

23.71 0.173 0.025 2.373 0.060 0.761 1.394 0.851 0.066 0.141 0.078 0.053 0.028 0.010 0.006

0.040 −0.008 0.037⁎⁎ 23,848 0.101

Note: Year of Survey (YOS) and Year of Birth (YOB) are the actual calendar years divided by 100. ⁎⁎ p b 0.001.

increased from 5255 in 1986, the initial child collection survey year, to a total of 11,504 in 2010, the last survey year we use in the current paper. The children were surveyed every two years, and at each survey they received a series of cognitive tests. In addition, the mothers were also interviewed, and information about the household was collected by the interviewer. The interviews of the children were typically conducted in the home of the child's mother by experienced, specially trained field staff. Child interviews through 1992 were conducted primarily in person using paper and pencil. Beginning in 1994, the interviews were administered using Computer-Assisted Personal Interviewing (CAPI). Spanish translations of several of the test instruments were made available to respondents with limited proficiency in English. However, the number of children who were assessed in Spanish was very low. For example, in 2000, the number of children assessed in Spanish was fewer than 10, and at this year, most of the Spanish language parents would have resided in the U.S. for more than two decades. 3.1.2. Variables and measurement 3.1.2.1. Intelligence. For the early cohort our measure of intelligence is derived from participants' test scores in the Armed Forces Qualifying Test (AFQT). This test was administered to groups of five to ten participants of the NLSY during the period of June through October 1980. Respondents were compensated, and the overall completion rate was 94%. The intelligence score in the NLSY is the sum of the standardized scores of four tests: arithmetic reasoning, paragraph comprehension, word knowledge and mathematics knowledge, and is expressed as on a standard intelligence scale of a mean of 100 and standard deviation of 15 scored on the basis of the US army scoring scheme aimed at achieving nationally representative standard scores (see addendum to attachment 106 of the NLSY). The reliability of the AFQT in our data is 0.92. The validity of the AFQT has been demonstrated in numerous studies including the prediction of training success (e.g., Ree & Earles, 1991) job performance (e.g., Scribner, Smith, Baldwin, & Phillips, 1986), as well as other measures of socio-economic success (see an extended discussion of the NLSY user guide6). For the later cohort our measure of intelligence in the NLSYC is The Peabody Picture Vocabulary Test (PPVT). This test “measures an individual's receptive (hearing) vocabulary for Standard American English and provides, at the same time, a quick estimate of verbal ability or scholastic aptitude” (Dunn & Dunn, 1981). The test is considered a good estimate for General Mental Ability (e.g., Campbell, Bell, & Keith, 2001; Childers, Durham, & Wilson, 1994; Snitz, Bieliauskas, Crossland, Basso, & Roper, 2000; Smith, Smith & Dobbs, 1991. But see Bracken & Prasse, 1982; Hodapp & Gerken, 1999). The PPVT was show to have a high reliability (e.g., Dunn & Dunn, 1981, report a median split reliability of 0.80) as well as high construct and 6 http://www.nlsinfo.org/nlsy79/docs/79html/codesup/ NLSY79%20Attachment%20106,%20Profiles%20of%20American%20Youth.pdf.

Please cite this article as: Ganzach, Y., Cognitive ability and party affiliation: The role of the formative years of political socialization, Intelligence (2017), http://dx.doi.org/10.1016/j.intell.2017.01.003

Y. Ganzach / Intelligence xxx (2017) xxx–xxx

5

Table 4 Descriptive statistics – Study 2.

DA Sex black White Education intelligence

Mean

STD

DA

Sex

Black

Whit

Educ.

Intel.

0.40/0.40 1.50/1.49 0.25/0.28 0.59/0.53 13.3/11.93 96.1/89.9

1.34/1.12 0.50/0.50 0.43/0.45 0.49/0.50 2.56/2.22 14.9/18.8

– 0.07 0.40 −0.37 −0.07 −0.25

0.09 – 0.00 0.00 0.06 0.00

0.34 0.00 – −0.70 −0.07 −0.35

−0.32 0.00 0.00 – 0.18 0.42

0.00 0.10 0.00 0.04 – 0.59

−0.13 0.01 −0.34 0.44 0.21 –

Means and standard deviations to the left (right) are of the NLSY79 (NLSYC). Correlations below (above) the diagonal are of the NLSY79 (NLSYC). For the NLSY79 (NLSYC) n varies between 7234 and 12,686 (4787 and 11,504).

predictive validity (Dunn & Dunn, 1981). The NLSYC includes other cognitive tests, but we chose this test since it provides the best estimation of intelligence among the tests that were available in the NLSY79C.7 The PPVT consists of 175 vocabulary items of generally increasing difficulty. The child listens to a word uttered by the interviewer and then selects one of four pictures that best describes the word's meaning. A child's entry point into the assessment is based on his or her age. In a few cases, a Spanish version of the PPVT-R was used until 2000. We used the standard age-normed scores of the PPVT which are provided by the NLSYC with a mean of 100 and a standard deviation of 15. Because of the structure of the NLSYC survey, children took the test at various ages, though an effort was made that most of them will take the test at age 10–11. Since the NLSYC contains repeat assessment of the same test, we averaged over the various assessments of the PPVT. The average number of assessments of this test was about 1.5 per child. 3.1.2.2. Party affiliation. Party affiliation is measured both in the NLSY79 and in the NLSYC on the basis of two questions. The first asks respondents to indicate whether they consider themselves Democrats, Republicans or Independents. If they answer Democrat [Republican], they were asked if they are strong Democrat [Republican] or not very strong Democrat [Republican]. Based on these two questions we constructed a 1–5 scale ranging from strong republican affiliation (1) independent (3) strong Democratic Affiliation (5). 3.2. Results and discussion Table 4 presents means, standard deviations and inter-correlations for the NLSY79 and the NLSYC. As in Study 1, the zero order correlation between intelligence and DA is negative in both studies (− 0.25 and − 0.13 for the NLSY79 and NLSYC, respectively). However, more importantly for the subjects of the current study, this correlation is less negative (or more positive) in the younger NLSYC sample than in the older NLSY79 sample (p b 0.0001). Table 5 presents the results of regression predicting DA from intelligence, race and education. (We do not present results of regressions that include controls for income since many of the participants in the NLSYC did not start their employment career). The results in this table indicate that intelligence is positively related to DA in the later cohort and negatively related to DA in the early cohort. These results are consistent with the cohort hypothesis. Note also that whereas the zero order correlation between intelligence and DA is negative (r = − 0.13, see Table 4), this relationship is positive when race and education are controlled for. These results are consistent with Ganzach's (2016) that suggested that adding race and socio-economic variables have a strongly effect on the observed relationship between intelligence and party affiliation, even reversing this observed relationship. 7 The other cognitive tests that are available in the NLSYC are the PIAT-math, PIAT-reading recognition, and PIAT- reading comprehension. These are all considered achievement tests.

4. Discussion Although most of the literature indicates a positive relationship between intelligence and liberal political preferences, there are some exceptions to this general finding. Studying the conditions under which these exceptions occur is important to understanding the development of political preferences. In the current paper we focus on time as a moderator of this relationship, and in particular on time as a moderator of the relationship between intelligence and Democratic Affiliation. We examine two hypotheses regarding this moderation, one in which these are the formative years in one's life that shape his or her current political preferences, and the other in which the period is which political preferences are elicited these preferences. Our results clearly support the first hypothesis. We rely on two methods in examining our hypotheses. One method is based on analysis of cross-sectional databases collected over a number of years. These data allow for separating the effect of period (Year of Survey) from the effect of formative years (Year of Birth), and in particular separating the effects of their interaction with intelligence. Note that the high correlation between YOS and YOB imposes restrictions on databases that can be used for this analysis. First, because the high multi-collinearity between YOS and YOB then need to include a large number of participants. Second, on the one hand they need to span not too many years so that the correlation between YOS and YOB will not be too high (which is the case if too many years are involved), but on the other hand, not too few years (to capture sufficient enough changes in processes determining political preferences). The other method that we use to examine our hypotheses is based on a comparison between cohorts at the same survey-time. In this comparison, YOS is being held constant (at 2008 in our study) so any time effect is associated with YOB. The two cohorts we use are particularly appropriate to examine our hypotheses, since, by virtue that the later cohort represents the children of the early cohort, differences cannot be explained by changes in the sample or population. Note, however, that although our studies distinguish between cohort and period effects, they do not distinguish these two effects from the effect of age. By construction, these three effects are always perfect linear functions of each other. As was noted by a number of authors, the effect of these three variable cannot be separated through statistical techniques, and a theoretical assumption is needed to separate them (Converse, 1976; Firebaugh, 1997; Glenn, 1989, 1994). Indeed, the theoretical assumption that underlies our analysis is that age is not relevant to party identity. Although this assumption is consistent with the literature, to the best of our knowledge there are no direct evidence to support it. Furthermore, although the current study identified clear time dependent trends in the intelligence-DA relationship, the processes underlying these trends remain unclear. These trends may reflect trends in the relationship between intelligence and liberal attitudes, associated with the intellectual zeitgeist, or they may reflect changes in the ideologies of the Democratic and Republican parties. Thus one direction for further research is to explore the effect of time on the relationship between intelligence and political attitudes rather than the effect of time

Please cite this article as: Ganzach, Y., Cognitive ability and party affiliation: The role of the formative years of political socialization, Intelligence (2017), http://dx.doi.org/10.1016/j.intell.2017.01.003

6

Y. Ganzach / Intelligence xxx (2017) xxx–xxx

Table 5 The effect of time on Democratic Affiliation – Study 2. NLSY79

NLSYC

Model 1 Intercept Sex Black White Education Intelligence n R2

Model 2

0.695 0.183⁎⁎ 0.775⁎⁎ −0.426⁎⁎

0.114 0.029 0.043 0.041

−0.006⁎⁎ 6895 0.192

0.001

0.660 0.176⁎⁎ 0.760⁎⁎ −0.423⁎⁎ 0.018⁎ −0.008⁎⁎

Model 1 0.114 0.029 0.043 0.041 0.007 0.001

6895 0.192

Model 2

−0.155 0.157⁎⁎ 0.538⁎⁎ −0.438⁎⁎

0.099 0.031 0.041 0.043

0.0034⁎⁎ 4690 0.137

0.0010

−0.256 0.150⁎⁎ 0.536⁎⁎ −0.437⁎⁎ 0.012 0.0029⁎

0.121 0.031 0.041 0.043 0.009 0.0010

4691 0.137

⁎ p b 0.01. ⁎⁎ p b 0.001.

on the intelligence-DA relationship. Meisenberg (2015) took a first step in this direction and reported the correlations between intelligence and liberal attitudes in various time periods in the GSS. These correlations do no exhibit any clear trend. However, Meisenberg's analysis should be viewed only as preliminary. Future analysis should use a regression based method, distinguishing between YOB and YOS, and take into account relevant control variables, such as education and income. In addition, the measurement of liberal beliefs should be expanded and include not only answers to a general question about liberal attitudes (a scale ranging from “Extremely liberal” to “Extremely conservative”) but should also include one's personal perception of liberalism vs. conservatism, and on the meaning of this dimension during the time the question was asked. The cultural mediation hypothesis (Dutton, 2013; Woodley, 2010; Woodley of Menie & Dunkel, 2015) is potentially a viable theoretical framework to understanding the time-dependent trends in the relationship between intelligence and political preferences. This hypothesis suggests that the more intelligent are better at detecting the dominant

cultural values in general and the dominant political preferences in particular. Thus our results may be explained by political preferences being more (less) democratic at earlier (later) cohorts. This suggest two testable hypothesis for future research: (1) the population average of political attitudes at the time of young adulthood are related to political attitudes at later years; and (2) intelligence moderates this relationship. It is obvious that the patterns of relationships between intelligence and party identity that were examined in the current study are specific to a particular country and particular period of time, and depend on the particular political circumstances of place and time. However, beyond providing some clarification regarding the relationship between intelligence and party affiliation in the US in recent decades, a broader lesson that can be learned from the current results is that the effect of intelligence on political preferences changes with time, and that the time which shapes this effect is primarily the formative years in which political preferences are developed. The current study does not identify the characteristics of the formative years that shape this effect. Identifying these characteristics is an interesting topic for future research.

Appendix 1. Democratic Affiliation and the time × intelligence interactions – Study 1 (ordered logit regression)

Model 1

Sex Black White Intelligence YOS YOB Income Education Income × YOS Income × YOB Education × YOS Education × YOB Intelligence × YOS Intelligence × YOB

Model 2

Model 3

b

stderr

b

stderr

b

stderr

0.184 0.952 −0.594 −0.716 0.392 −5.106

0.022 0.059 0.053 0.132 0.764 0.414

0.178 0.877 −0.627 −0.097 −0.019 −0.862 −0.168 −5.288

0.023 0.063 0.056 0.013 0.005 0.140 0.817 0.451

−0.009 0.046⁎⁎

0.008 0.004

−0.0037 0.0480⁎⁎

0.0080 0.0045

0.173 0.884 −0.626 9.729 −5.921 −0.491 4.577 −7.071 −0.695⁎⁎ 0.206⁎⁎ 0.239⁎⁎ 0.058 −0.013 0.039⁎⁎

0.023 0.063 0.056 2.228 0.798 0.161 1.310 0.719 0.133 0.073 0.049 0.027 0.009 0.005

Note: Year of Survey (YOS) and Year of Birth (YOB) are the actual calendar years divided by 100. ⁎⁎p b 0.001.

Appendix 2. The time × intelligence interactions by race

Blacks Intercept Sex Income Education Intelligence YOS YOB

b 33.767 0.172 2.024 −2.687 −0.041 3.490 −5.095

Whites stderr 44.878 0.053 4.381 1.793 0.385 2.659 1.530

b 64.76 0.185 13.011 −6.150 −0.871 0.992 −4.054

stderr 29.25 0.029 2.852 1.009 0.206 1.692 0.918

Please cite this article as: Ganzach, Y., Cognitive ability and party affiliation: The role of the formative years of political socialization, Intelligence (2017), http://dx.doi.org/10.1016/j.intell.2017.01.003

Y. Ganzach / Intelligence xxx (2017) xxx–xxx

7

Appendix 2 (continued) (continued) Blacks Income × YOS Income × YOB Education × YOS Education × YOB Intelligence × YOS Intelligence × YOB n R2

0.119 −0.223 0.086 0.052 −0.049 0.053⁎⁎ 3405 0.048

Whites 0.266 0.155 0.113 0.057 0.023 0.013

−0.706 0.046 0.260 0.048 0.015 0.029⁎⁎ 19,356 0.028

0.167 0.092 0.061 0.033 0.012 0.006

Note: Year of Survey (YOS) and Year of Birth (YOB) are the actual calendar years divided by 100. ⁎⁎p b 0.001.

References Abramowitz, A. I., & Saunders, K. L. (1998). Ideological realignment in the US electorate. The Journal of Politics, 60(3), 634–652. Alwin, D. F., Cohen, R. L., & Newcomb, T. M. (1991). Political attitudes over the life span: The Bennington women after fifty years. University of Wisconsin Press. Bracken, B. A., & Prasse, D. P. (1982). Comparison of the PPVT, PPVT-R, and intelligence tests used for the placement of black, white and hispanic EMR students. Journal of School Psychology, 19(4), 304–311. Campbell, J. M., Bell, S. K., & Keith, L. K. (2001). Concurrent validity of the peabody picture vocabulary test-as an intelligence and achievement screener for low SES African American children. Assessment, 8(1), 85–94. Carl, N. (2014a). Verbal intelligence is correlated with socially and economically liberal beliefs. Intelligence, 44, 142–148. Carl, N. (2014b). Cognitive ability and party identity in the United States. Intelligence, Intelligence, 47, 3–9. Childers, J. S., Durham, T. W., & Wilson, S. (1994). Relation of performance on the Kaufman Brief Intelligence Test with the Peabody Picture Vocabulary Test—Revised among preschool children. Perceptual and Motor Skills, 79(3), 1195–1199. Converse, P. E. (1976). The dynamics of party support: Cohort analyzing of party identification. Beverly Hills, CA: Sage. Deary, I. J., Batty, G. D., & Gale, C. R. (2008a). Bright children become enlightened adults. Psychological Science, 19, 1–6. Deary, I. J., Batty, G. D., & Gale, C. R. (2008b). Childhood intelligence predicts voter turnout, voting preferences, and political involvement in adulthood: The 1970 British Cohort Study. Intelligence, 36(6), 548–555. Dunn, L. M., & Dunn, L. M. (1981). PPVT-R Revised Manual. Circle Pines. Minnesota: American Guidance Service, Inc. Dutton, E. (2013). The cultural mediation hypothesis: A critical examination. Intelligence, 41(5), 321–327. Firebaugh, G. (1997). Analyzing repeated surveys. Sage university paper series on quantitative applications in the social sciences, No. 07–115Thousand Oaks, CA: Sage. Ganzach, Y. (2016). Cognitive ability and party identity: No important differences between democrats and republicans. Intelligence, 58, 18–21. Glenn, N. D. (1989). A caution about mechanical so-lutions to the identification problem in cohort analysis: Comment on Sasaki and Suzuki. American Journal of Sociology, 95, 754–761. Glenn, N. D. (1994). Television watching, newspaper reading, and cohort differences in verbal ability. Sociology of Education, 67, 216–230. Hauser, R. M., & Huang, M. H. (1997). Verbal ability and socioeconomic success: A trend analysis. Social Science Research, 26(3), 331–376. Hess, E. H. (1973). Imprinting. New York: Van Nostrand. Heaven, P. C. L., Ciarrochi, J., & Leeson, P. (2011). Cognitive ability, right-wing authoritarianism, and social dominance orientation: A five-year longitudinal study amongst adolescents. Intelligence, 39, 15–21.

Hodapp, A. F., & Gerken, K. C. (1999). Correlations between scores for Peabody Picture Vocabulary Test—III and the Wechsler Intelligence Scale for Children—III. Psychological Reports, 84(3 suppl), 1139–1142. Hodson, G., & Busseri, M. A. (2012). Bright minds and dark attitudes: Lower cognitive ability predicts greater prejudice through right-wing ideology and low inter-group contact. Psychological Science, 23, 187–195. Kanazawa, S. (2004). General intelligence as a domain-specific adaptation. Psychological Review, 111(2), 512. Kanazawa, S. (2010). Why liberals and atheists are more intelligent. Social Psychology Quarterly, 73, 33–57. Karadja, M., Mollerstrom, J., & Seim, D. (2014). Richer (and holier) than thou? The effect of relative income improvements on demand for redistribution. The Review of Economics and Statistics. Layman, G. C., & Carsey, T. M. (2002). Party polarization and “conflict extension” in the American electorate. American Journal of Political Science, 786–802. Levendusky, M. (2009). The partisan sort: How liberals became democrats and conservatives became republicans. University of Chicago Press. Meisenberg, G. (2015). Verbal ability as a predictor of political preferences in the United States, 1974–2012. Intelligence, 50, 135–143. Miller, G., & Schofield, N. (2008). The transformation of the republican and democratic party coalitions in the US. Perspectives on Politics, 6(3), 433–450. Peterson, S. A. (1983). Biology and political socialization: A cognitive developmental link? Political Psychology, 265–288. Pesta, B., & McDaniel, M. A. (2014). State IQ, well-being and racial composition as predictors of U.S. presidential election outcomes. Intelligence, 42, 107–114. Pesta, B., McDaniel, M. A., & Bertsch, S. (2010). Toward an index of well-being for the fifty states. Intelligence, 38, 160–168. Ree, M. J., & Earles, J. A. (1991). Predicting training success: Not much more than g. Personnel psychology, 44(2), 321–332. Schoon, I., Cheng, H., Gale, C. R., Batty, G. D., & Deary, I. J. (2010). Social status, cognitive ability, and educational attainment as predictors of liberal social attitudes and political trust. Intelligence, 38, 144–150. Scribner, B. L., Smith, D. A., Baldwin, R. H., & Phillips, R. L. (1986). Are smart tankers better? AFQT and military A productivity. Armed Forces & Society, 12(2), 193–206. Stankov, L. (2009). Conservatism and cognitive ability. Intelligence, 37, 294–304. Smith, T. C., Smith, B. L., & Dobbs, K. (1991). Relationship between the peabody picture vocabulary test-revised, wide range achievement test-tevised, and weschler intelligence scale for children-revised. Journal of School Psychology, 29(1), 53–56. Snitz, B. E., Bieliauskas, L. A., Crossland, A., Basso, M. R., & Roper, B. (2000). PPVT-R as an estimate of premorbid intelligence in older adults. The Clinical Neuropsychologist, 14(2), 181–186. Woodley, M. A. (2010). Are high-IQ individuals deficient in common sense? A critical examination of the ‘clever sillies’ hypothesis. Intelligence, 38, 471–480. Woodley of Menie, & Dunkel, C. S. (2015). Beyond the cultural mediation hypothesis: A reply to Dutton (2013). Intelligence, 49, 186–191.

Please cite this article as: Ganzach, Y., Cognitive ability and party affiliation: The role of the formative years of political socialization, Intelligence (2017), http://dx.doi.org/10.1016/j.intell.2017.01.003