An Analysis of the Public’s
Impressions of Labor Unions: Scale Validation and Application
Gordon L. Freeman, Jr.
H. Ronald Moser
MiddleTennesseeStateUniversity
Introduction
General Overview
According to information recently released by the Federal Bureau of Labor Statistics, 2004 was not a good year for union membership organizing efforts. The percentage of U.S. workers belonging to a union decreased from 12.9 percent in 2003 to 12.5 percent in 2004. The union membership rate has steadily declined from a high of almost 35 percent in 1960. The decline has been continuous despite the fact that in recent years AFL-CIO unions have been spending more than $40 million a year on worker-organizing activities. The decline has been the subject of numerous scholarly studies analyzing the causes and offering possible solutions (Dobbs, 2005). None of the published research has identified a single cause, but most writers agree that the reasons can be grouped into five categories: (1) increased individualism of workers and their attitude toward unions, (2) legal support for unions and existence of right-to-work laws, (3) macroeconomic factors, (4) employer opposition initiatives, and (5) union officials’ opposition to representation elections (Atleson, 1994; Jarley and Kuruville, 1994; Karier, 1991; Leap, 1990; Margalioth, 1998; Saporta and Lincoln, 1995). However, in a 2004 nationwide Zogby poll, 63 percent of the respondents answered that, in general, they approve of labor unions.
Some researchers have argued that right-to-work laws actually have little impact on the percentage of workers joining unions. Others argue that the existence of such laws is really a reflection of local attitudes toward unions. They believe that primarily the attitude of local workers affects their decision to join a union. A comprehensive study analyzing many articles on the topic concluded that the impact of right-to-work laws remains significant but of a smaller magnitude than previously thought (Davis and Huston, 1993).
Purposes of the Study
The purposes of this study were to determine (a) a factor score that represents the public’s impressions of labor unions and (b) whether any of the various demographic variables accounted for any significant difference in the public’s impressions of labor unions.
The Present Study
We will use a survey instrument previously developed by the University of Michigan Survey Research Center (Quinn and Staines, 1978) to investigate the impressions the public has of labor unions. In the process, we will refine and validate the Public’s Impressions of Labor Unions Scale (PILUS) using principal component factor analysis. We will divide the scale values into three categories (high, median, and low) corresponding to the respective impressions of labor unions, and we will perform cross-tabulations on these scale categories and various demographic variables to see how impressions differ among the demographic sections. Using multivariate analysis of covariance (MANCOVA) on the PILUS scores and various control variables, we will ascertain whether there are any statistically significant differences in people’s impressions of labor unions for those statements that are not included in calculating PILUS scores.
During exploratory factor analysis we selected 11 of the 18 statements reflecting the public’s impressions of labor unions to use in the calculation of PILUS scores. Do these PILUS scores also represent differences in the responses to the seven statements that were not used in the calculation of the PILUS scores? Based on the results of statistical tests, we were able to use MANCOVA to test six of the seven statements that were not used in the calculation of the PILUS scores. The statement “Most labor unions are benevolent,” when included in the MANCOVA, provided too high a level of interaction to conduct a valid test with a 95 percent level of confidence. As a result, we could only test the following hypotheses.
Hypothesis 1: People with high Public’s Impressions of Labor Union Scale (PILUS) scores are more likely than those with median and low PILUS scores to agree that labor unions are democratic and necessary in most companies.
Hypothesis 2: People with high Public’s Impressions of Labor Union Scales (PILUS) scores are more likely than those with median and low PILUS scores to disagree with statements that labor union dues and fees are too high, that labor unions are useless, hold back the progress of business, and force people to join who do not want to join.
Methodology
We adopted and modified the research instruments developed by the University of Michigan Survey Research Center (Quinn and Staines, 1978). The survey questionnaire examines respondents’ demographic characteristics (age, race, gender, size of place of employment, employment status, education, income from primary job, union membership, parents’ union membership, industry, and job satisfaction). This instrument has an 18-item scale, presented in Table 1, designed to assess the respondents’ general impressions of labor unions. All of the questions, though slightly modified, were selected from questions that have been included in various published opinion polls conducted over the past 30 years. We used a Likert-type scale to designate agree (3), undecided (2), and disagree (1).
The data for this study were obtained through mail questionnaires sent to a random sample of 6,000 residents from telephone books from the seven metropolitan statistical areas in Tennessee: Memphis, Nashville, Clarksville, Chattanooga, Knoxville, Jackson, and Tri-Cities (Bristol, Kingsport, and Johnson City). Appropriate numbers of residents from each city were drawn according to the ratio of each city’s population to the combined population of all seven urban areas. The research instrument was mailed to these residents, and 669 usable questionnaires were received and used in this study. The return rate was 11.2 percent.
A confidence level of 95 percent (i.e., a 0.05 level of significance) is traditionally used for consumer research projects (Lind, Marchal, and Wathen, 2005). The size of the random samples taken and the number of usable responses received was sufficient to produce a confidence level of at least 95 percent in the results of each statistical procedure performed.
Demographic Variables
Among the 669 participants, 143 were female, and 468 were male (data not available = 58). Regarding age, 366 (54.7%) were 45 or younger, 298 (44.5%) were over 45, and 5 (0.8%) did not specify age. In terms of race, the majority were white (82.5%). Regarding income from primary job, 104 (15.5%) earned $15,000 or less, 429 (64.1%) earned $15,001-45,000, 109 (16.3%) earned $45,001 or more, and 27 (4.0%) did not specify family income. In terms of education, 358 (53.5%) had no college degree, 303 (45.3%) had a college degree or advanced degree, and 8 (1.2%) did not specify level of education. Participants in the study were mostly male, 45 or under, and white with less than a college degree and low to middle income. Therefore, participants in this sample do not match the population perfectly. However, there was no reason to believe this sample was atypical.
Findings
Exploratory Factor Analysis
The survey contained 18 statements that described general impressions of labor unions. The values entered for these 18 statements were coded so that a higher value reflected a more positive impression of unions. A test for internal reliability was performed using SPSS. This test, applied to the responses to the 18 statements used to derive common factors, revealed a Cronbach’s alpha of 0.937, well above the minimum acceptable value of 0.70, indicating that this set of data has a high level of internal reliability. A random sample of 336 (approximately 50 percent) was selected from the survey data, and the 18 items were analyzed using SPSS. Because 26 cases contained missing values, the analysis was performed for n = 310 cases. Two factors exceeded the minimum eigenvalue of one and were verified using a scree test. Using two as the number of factors, the principal component method of factor analysis was repeated on the sample, this time with a varimax rotation. The two significant factors identified were Factor 1, positive impressions of unions (9 items, eigenvalue = 9.122, % of variance = 50.7), and Factor 2, negative impressions toward unions (9 items, eigenvalue = 1.559, % of variance = 8.7). The two factors altogether explained 59% of the variance. However, the names given to these factors describe only the predominant classification of the items associated with each factor, and many items had high cross-loadings.
By eliminating the items with high cross-loadings, we selected 11 items for additional study. Using the random sample (n = 321) with the 11 items, we clearly identified the two factors previously described. A test for internal reliability on these 11 items, using the full sample, was performed using SPSS. It revealed a Cronbach’s alpha of 0.916, indicating a high level of internal reliability. The two-factor solution, obtained using the random sample, was validated by applying a factor analysis to the full sample. The score created by the sum of the two factor scores was labeled the PILUS score. Table 2 shows the individual items, their loadings on each factor, their total rotation sums of squared loadings, and explained variance (68 percent) for the full sample.
Cross-Tabulations
The scores were recoded as low, median, and high for a cross-tabulation analysis on the scores against the various demographic variables. Categories with expected frequencies below five were eliminated from the analysis. The seven significant results, shown in Tables 3-9, were for race, education, income, union membership, parent union membership, industry, and job satisfaction. Although blacks represented about 10 percent of the 669 respondents, they accounted for more than 20 percent of the high PILUS scores. Almost 40 percent of the high PILUS scores were from respondents whose education included no college, but this group represented less than 24 percent of the 669 respondents. More than 16 percent of the 669 respondents have an income from their primary job higher than $45,000, but this group represented only eight percent of the high PILUS scores. Almost 38 percent of the high PILUS scores were from union members, but union members represented less than 16 percent of the 669 respondents. While less than 23 percent of the 669 respondents had one or more parents who were union members, this group accounted for almost 31 percent of the high PILUS scores. Almost 10 percent of the 669 respondents were employed in retail trade, but this group accounted for less than six percent of the high PILUS scores. Almost three percent of the 669 respondents were employed in banking and finance, but this group accounted for only about one percent of the high PILUS scores. More than four percent of the 669 respondents were employed in wholesale trade, but this group accounted for less than two percent of the high PILUS scores. In contrast, a little more than four percent of the 669 respondents were employed by the federal government, but this group accounted for more than 10 percent of the high PILUS scores. Similarly, transportation accounted for less than six percent of the 669 respondents, but transportation accounted for more than 10 percent of the high PILUS scores. Finally, just over seven percent of the 669 respondents were not very satisfied with their employment, but this group represented almost 13 percent of the high PILUS scores.
Multivariate Analysis of Covariance Results
A multivariate analysis of covariance (MANCOVA) was applied to the three-way split of the PILUS scores (low, median, and high) to determine how the PILUS scores related to the two items “Labor unions are necessary in most companies” and “Most labor unions are democratic.” The MANCOVA was performed using race, education, income, union membership, parent union membership, industry, and job satisfaction as control variables. Summary results are listed in Table 10. Because Box’s Test of Covariance Matrices produced a significant result (P-value < 0.001), Pillai’s Trace was used instead of Wilks’ Lambda (Mertler & Vannatta, 2005). The test for interaction between the factor and the covariates did not produce a significant result [F(6, 848) = 2.058, Pillai’s Trace = 0.029, P-value = 0.056]. The overall MANCOVA was significant [F(4, 854) = 59.060, Pillai’s Trace = .433, P-value < 0.001, partial Eta squared = 0.217]; that is, significant differences in these two items were found among the three levels of PILUS scores. Univariate ANOVA results reveal that PILUS levels differ for the necessary statement [F(2, 427) = 111.336, P-value < 0.001, partial Eta squared = 0.343] and the democratic statement [F(2, 427) = 64.949, P-value < 0.001, partial Eta squared = 0.233]. That is, the three levels of PILUS scores had different impressions regarding the necessity of labor unions in most companies and whether labor unions are democratic. Helmert contrasts indicated that respondents with high PILUS scores were significantly more likely to agree both that labor unions are necessary in most companies (mean difference = 0.966, P-value < 0.001) and that labor unions are democratic (mean difference = 0.871, P-value < 0.001) than those with median and low PILUS scores. Furthermore, the median PILUS group was significantly higher than the low PILUS group on both statements. Consequently, Hypothesis 1 was supported.
A multivariate analysis of covariance (MANCOVA) was used to determine how the PILUS scores related to the four items “Labor unions force people to join who do not want to join,” “Labor unions hold back progress of businesses,” “Most labor union dues and fees are too high,” and “Labor unions are useless.” The MANCOVA was performed using race, education, income, union membership, parent union membership, industry, and job satisfaction as control variables. Summary results are listed in the second portion of Table 10. Because Box’s Test of Covariance Matrices produced a significant result (P-value < 0.001), Pillai’s Trace was used instead of Wilks’ Lambda (Mertler & Vannatta, 2005). The test for interaction between the factor and the covariates did not produce a significant result [F(12, 1257) = 1.213, Pillai’s Trace = 0.034, P-value = 0.268]. The overall MANCOVA was significant [F(8, 842) = 42.875, Pillai’s Trace = .579, P-value < 0.001, partial Eta squared = 0.289]; that is, the three levels of PILUS scores had different impressions regarding the four statements “Labor unions force people to join who do not want to join,” “Labor unions hold back progress of businesses,” “Most labor union dues and fees are too high,” and “Labor unions are useless.” Univariate ANOVA results reveal that PILUS levels differ for the “force people” statement [F(2, 423) = 43.946, P-value < 0.001, partial Eta squared = 0.172], the “hold back progress” statement [F(2, 423) = 78.013, P-value < 0.001, partial Eta squared = 0.269], the “fees are too high” statement [F(2, 423) = 27.431, P-value < 0.001, partial Eta squared = 0.115], and the “useless” statement [F(2, 427) = 158.745, P-value < 0.001, partial Eta squared = 0.429]. That is, the three levels of PILUS scores had different impressions regarding each of the four statements “Labor unions force people to join who do not want to join,” “Labor unions hold back progress of businesses,” “Most labor union dues and fees are too high,” and “Labor unions are useless.” Helmert contrasts among the three groups revealed that the high PILUS group was significantly higher than the median and low PILUS groups in terms of impressions of all four statements. Because these four items were coded as 3 = Disagree, 2 = Undecided, and 1 = Agree, this result demonstrates that the high PILUS group is more likely to disagree with these four statements than the median and low PILUS groups. Furthermore, the median PILUS group was significantly higher than the low PILUS group on all four statements. Based on these results, Hypothesis 2 is supported, and the PILUS measure is validated.
Discussion
In this study, we identified two factors for calculating the PILUS scores using the principal component procedure on a random sample taken from the full sample. The 11-item PILUS that was derived had high internal reliability (Cronbach’s alpha = 0.916) when applied to the full sample. The PILUS scores were then recoded into high, median, and low groups, whereby the group with high PILUS scores corresponded with a more positive impression of labor unions than the groups with median or low PILUS scores. A cross-tabulation of the PILUS groups against various demographic variables showed a significant result for race, education, income, union membership, parent union membership, industry, and job satisfaction. The following paragraph offers some general statements based on examination of this cross-tabulation analysis.
In general, blacks have a more positive impression of labor unions than whites. The less educated group (no college) has a more positive impression of labor unions than the higher educated group (at least some college). People with higher incomes do not have as positive an impression of labor unions as people with lower incomes. Expectedly, union members have a more positive impression of labor unions than nonmembers. While the difference is not as pronounced, a person with at least one parent who was a member of a labor union views labor unions more positively than one whose parents do not have such an affiliation. People employed in retail trade, banking and finance, or wholesale trade tend to view labor unions less positively than people in most other industries, whereas people employed by the federal government or the transportation industry tend to view labor unions more positively than those in other industries. People who are not very satisfied with their employment tend to view labor unions more positively than people who are satisfied to some extent with their employment.
Based on the results of statistical tests, we were able to use MANCOVA to test six of the seven statements that were not used in the calculation of the PILUS scores. The statement “Most labor unions are benevolent,” when included in the MANCOVA, provided too high a level of interaction to conduct a valid test with a 95 percent level of confidence. The responses to the remaining six statements (not used in the calculation of the PILUS scores) were tested using MANCOVA to determine the level to which the responses to these six statements differed depending on the level of the PILUS score. Two MANCOVA analyses were performed. The first analysis was on the two statements expressing a positive impression of labor unions, and the second analysis was on the four statements expressing a negative impression of labor unions.
The first MANCOVA analysis was applied to the three-way split of the PILUS scores (low, median, and high) to determine how the PILUS scores related to the two items “Labor unions are necessary in most companies” and “Most labor unions are democratic.” This analysis determined that the three levels of PILUS scores had different impressions regarding the necessity of labor unions in most companies and whether labor unions are democratic. Helmert contrasts indicated that respondents with high PILUS scores were significantly more likely to agree both that labor unions are necessary in most companies and that labor unions are democratic than those with median and low PILUS scores. Furthermore, the median PILUS group was significantly higher than the low PILUS group on both statements. These results suggest that the three levels of PILUS scores generally agree with the responses to the two items “Labor unions are necessary in most companies” and “Most labor unions are democratic,” giving further evidence that the PILUS scores accurately reflect the public’s impressions of labor unions.
The second MANCOVA analysis was used to determine how the three-way split of the PILUS scores (low, median, and high) related to the four items “Labor unions force people to join who do not want to join,” “Labor unions hold back progress of businesses,” “Most labor union dues and fees are too high,” and “Labor unions are useless.” This analysis determined the three levels of PILUS scores had different impressions regarding each of the four statements “Labor unions force people to join who do not want to join,” “Labor unions hold back progress of businesses,” “Most labor union dues and fees are too high,” and “Labor unions are useless.” Helmert contrasts among the three groups revealed that the high PILUS group was significantly higher than the median and low PILUS groups in terms of its impressions of all four statements. Because these four items were coded as 3 = Disagree, 2 = Undecided, and 1 = Agree, this result demonstrates that the high PILUS group is more likely to disagree with these four statements than the median and low PILUS groups. Furthermore, the median PILUS group was significantly higher than the low PILUS group on all four statements. These results suggest that the three levels of PILUS scores generally disagree with the responses to the four items “Labor unions force people to join who do not want to join,” “Labor unions hold back progress of businesses,” “Most labor union dues and fees are too high,” and “Labor unions are useless,” giving further evidence that the PILUS scores accurately reflect the public’s impressions of labor unions.
The study provides several contributions. First, we refined and validated the Public’s Impressions of Labor Unions Scale (PILUS). Researchers should find this scale useful in measuring people’s impressions of labor unions. Second, our study has identified segments of the public that have a more positive impression of labor unions than their percentage membership in the sample suggests. Third, our study has identified segments of the public that have a more negative impression of labor unions than their percentage membership in the sample suggests.
This study is not without its limitations. The return rate for the survey was low but acceptable (n = 669). The respondents in this study were located in a state in the southeastern U.S. and were mostly male, 45 or under, and white with less than a college degree and low to middle income. The respondents may have a positive bias toward research using survey questionnaires. This study may have common method variance since it is a cross-sectional survey of participants at one particular time without any follow-up. Because Tennessee is a right-to-work state, generalizability is expected to extend only to the public in other right-to-work states.
References
Atleson, James B. (1994). “Commentary: Law and Union Power: Thoughts on the United States and Canada,”
42 Buffalo L. Rev. 463.
Bureau of Labor Statistics. (2004, January). Union Membership (Annual), U.S. Department of Labor, URL:
http://stats.bls.gov/news.release/union2.toc.htm.
Davis, Joe C., & John H. Huston. (1993, January). “Right-to-Work Laws and Free Riding,” Economic Inquiry,
31, 52-59.
Dobbs, Lou. (March 2005). “Disorganized Labor,” U.S. News & World Report, 7: 48.
Jarley, Paul, & Sarosh Kuruvilla. (1994, Spring). “American Trade Unions and Public Approval:
Can Unions Please All of the People All of the Time?” Journal of Labor Research, XV, 97-116.
Karier, Thomas. (1991, February). “Accounting for the Decline in Private Sector Unionization Representation
Elections, Structural Changes and Restructuring,” Jerome Levy Institute Working Paper No. 44.
Leap, Terry L., William H. Hendrix, R. Stephen Cantrell, & G. Stephen Taylor. (1991).
“Discrimination against Prounion Job Applicants,” Industrial Relations: A Journal of Economy and Society,
29, 469-478.
Lind, Douglas A., William G. Marchal, & Samuel A. Wathen. (2005). Statistical Techniques in Business & Economics.
New York: McGraw-Hill/Irwin.
Margalioth, Sharon Rabin. (1998, Fall). “The Significance of Worker Attitudes: Individualism As a Cause for
Labor’s Decline,” Hofstra Labor and Employment Law Journal, 16, 133-166.
Mertler, Craig A., & Rachel A.Vannatta (2005). Advanced and Multivariate Statistical Methods. (3rd ed.).
Glendale, CA: Pyrczak.
Quinn, Robert P., & Graham L. Staines. (1978). The Quality of Employment Survey. University of Michigan,
Survey Research Center.
Saporta, Ishak, & Bryan Lincoln. (1995, Summer). “Managers’ and Workers’ Attitudes towards Unions in the
U.S. and Canada,” Industrial Relations, 50, 550-567.
|
