Discussion on White Southern support for the Confederate battle flag

Continuing from this Twitter thread...

Hi Logan,

1. I do not dispute the claimed correlation between White Southerners' racial attitudes and support for the Confederate battle flag, but the Wright and Esses 2017 analysis suggests an important causal claim that a meaningfully-large percentage of White Southerners support use of the Confederate battle flag for reasons unrelated to racial animus. Is there evidence that that causal claim is not correct?

2. I think that a "'heritage' doesn't tell us much" claim should be based on the performance of measures of pride in Southern heritage. Civil War knowledge and linked fate with Southerners are not measures of pride, so these measures cannot support a claim about the low explanatory power of pride.

3. Could you articulate what is inadequate about the Wright and Esses racial attitude measures? Given the results that Carney and Enos reported here, racial resentment does not appear to be an adequate measure of racial attitudes.

5 Comments on “Discussion on White Southern support for the Confederate battle flag

  1. Logan Strother responded to my above discussion here: https://loganstrother.wordpress.com/blog/. My reply to that discussion is below.

    ---

    Hi Logan,

    I'll address my main criticism first, which is that your analysis does not support the inference that pride in Southern heritage has low explanatory power, because the analysis includes neither a measure of such pride nor evidence that the included measures associate enough with pride in Southern heritage to be an informative proxy.

    My interpretation of your argument is that it is possible to make an inference about the low explanatory power of pride in Southern heritage because, relative to other persons, persons with such pride should [A] know more about the Civil War and [B] express more support for use of the Confederate flag, so that the observed negative association between [A] and [B] suggests that support for use of the flag cannot be explained by pride in Southern heritage.

    It's reasonable to expect persons with pride in Southern heritage to perform better on a test of Civil War knowledge than all-else-equal persons who lack such pride, but the data in your study do not permit an all-else-equal analysis that controls for factors such as cognitive ability and interest in U.S. history that can explain performance on a Civil War knowledge test. So for "why—if pride in southern heritage is racially innocuous—is knowledge of southern heritage negatively correlated with support for the chief symbol of southern heritage?", it could be that high-school-type knowledge or the more general ability to remember high-school-type facts or curiosity about such facts is lower among flag supporters than among other persons.

    In your Georgia dataset, controlling for sex, age, and education, Whites who preferred the Georgia state flag with the Confederate battle emblem scored 12 percentage points lower on the Civil War knowledge test than Whites who preferred some other flag. But it's possible that the relative performance of Whites who preferred the flag with the Confederate battle emblem would have been equally bad or worse on a history test about topics unrelated to the South.

    Three notes:

    * I realize that your article indicates that the positive correlation between [A] and [B] should appear controlling for education, but education is an insufficient statistical control because of the substantial range of history knowledge within groups of persons with the same education level. It's unreasonable to expect all persons with the same education level to perform similarly well on a test of history knowledge.

    * If persons with pride in Southern heritage on average know less about the Civil War all-else-equal, that group-level association would not necessarily mean that there is not a meaningfully-large percentage of White Southerners who support use of the Confederate battle flag for reasons unrelated to racial animus.

    * The negative association between [A] and [B] would also be expected if learning about the Civil War tends to cause a reduction of support for use of the Confederate flag.

    ---

    By the way, could you place online the full original Georgia dataset, with codebook? The replication dataset is missing items, such as the original items used for the measure for Civil War knowledge. Moreover, the Appendix reports that the introduction to the Sherman item is "Next, here are a few questions about the history of the state of Georgia", but the Appendix does not appear to report any further items about Georgia history. Did the original survey include other Georgia history items that are not reported in the Appendix?

    Regarding the Civil War knowledge test, the Appendix description of the Georgia survey reports the battle item wording as "Do you know the names of any Civil War battles?". You code participants as 0, 1, or 2 on this item, depending on how many battles they named. But were participants instructed to name multiple battles if they could?

    The Appendix reports the Sherman item as ending with "...Do you happen to know who William Tecumseh Sherman was?", with response options of Yes, No, and Can't recall. For participants who responded "Yes", did the survey verify whether the participant knew who Sherman was?

    ---

    Numbers below refer to numbers from the earlier discussion.

    1. I think that it's correct to not place much trust in point estimates from MTurk studies, which is why I asked about a meaningfully-large percentage of White Southerners and not about the majority suggested in Wright and Esses 2017.

    2A. Regarding the pride/flag association being limited to Whites, the 95% confidence interval for Southern pride in Wright and Esses 2017 Table 2 is [-0.062, 0.579] for Black participants, which is not statistically different from the [0.239, 0.425] estimate for White participants and for which the confidence interval end falls close enough to zero that I would not interpret the lack of statistical significance as evidence of a lack of an association.

    2B. Regarding the claim that "prejudice toward blacks is positively associated with warmth toward southerners, especially among whites who live in the south", the results reported in Table 3 of your article [i] concern racial resentment, which is not a good measure of racial prejudice (see the Carney and Enos paper), and [ii] concern a difference in feeling thermometer ratings, which ignores the information in the absolute values of the scales and thus does not differentiate between rating differences caused by relatively warm ratings of Whites and equivalent differences caused by cold ratings of Blacks. My analyses indicated that fewer than 10 percent of Southern participants in the 2008 ANES who rated Southerners and Blacks on the feeling thermometers rated Southerners at or above 50 and rated Blacks below 50, so about 90 percent of Southern participants for the Table 3 correlation had either non-cold ratings of Blacks or had cold ratings of Blacks and of Southerners. For a study presented as addressing the debate in which the relevant racial attitude is described as hate, I think that it's best to not infer much from an aggregate analysis in which the vast majority of persons rate Blacks at or above the midpoint of a feeling thermometer. Moreover, the strong within-group Southerner/Black-White-difference thermometer association does not mean that this association is present among all or even most individuals in that group.

    3. Wright and Esses 2017 have seven racial attitude measures, so I'm not sure why the "disturbing" item would need to capture contemporary racial prejudice. I think that a claim that the Wright and Esses 2017 measures are inadequate would require identifying an aspect of racial attitudes that is absent from their analysis and explaining why that aspect is needed. I'll stipulate that I think that this can be accomplished.

    ---

    Let me know if there is anything that I did not address that I should have addressed.

  2. Logan responded to my prior comment as an update on his blog: https://loganstrother.wordpress.com/blog/. My response is below:

    ---

    Hi Logan,

    4. I'd like to repeat my request that the full codebook and dataset for the Georgia survey be placed online and for you to indicate [1] whether there is evidence that participants who said they knew who Sherman was really did know who Sherman was and [2] whether participants were instructed to name multiple Civil War battles if they could (such that your analysis did not code a participant as knowing only one battle when the lack of a second battle offering was only because the participant did not know that they should have offered more knowledge).

    I'm especially interested in whether the Georgia survey included history items that are not reported in the Appendix of your article. If there are other history items and these items reflect what a person with Southern pride should know, then responses to these items could be added to the knowledge test to make the test more of a Southern heritage knowledge test. If there are other history items and these items are unrelated to what a person with Southern pride should know, then responses to these items could serve as a control for participant general history knowledge.

    ---

    5. I think that it's correct that a rating of Whites 90 / Blacks 60 on a feeling thermometer is meaningful. But I don't think that the meaning is hatred of Blacks, and I think that the meaning in a 90/60 rating differs from the meaning in a Whites 60 / Blacks 30 rating.

    I'm trying to imagine the ways in which anti-Black sentiment can cause a person to support use of the Confederate battle flag. I can imagine a person supporting use of the Confederate battle flag because this use would be harmful to Blacks in some way, or because this use would reflect the person's belief that Blacks should be treated as second-class citizens. These aren't the types of people that I expect to rate Blacks at 60 on a feeling thermometer, but I can much more easily imagine these people rating Blacks at 30.

    ---

    6. Regarding your comment that "... if this pride is neither racist nor rooted in appreciation for Southern Civil War history...", I don't think that your study's Civil War knowledge result indicates that persons who support the flag do not appreciate Civil War history, if "appreciate" means "place value on". Moreover, the claim seems to be contradicted by data from the Fall 1994 Southern Focus poll, which indicates that, in the presence of controls for demographics and a support-for-segregation item, disagreement that there should be no official use of the Confederate flag positively associated with agreement that "I have a great deal of interest in the history of the South" and positively associated with disagreement that "It's important to remember our history, but the Civil War doesn't mean much to me personally".

    Disagreement that there should be no official use of the Confederate flag did not associate at p<0.05 with the "name a Civil War battle" item, coded 1 if the participant named any Civil War battle and 0 otherwise, as best as I could code the item. The "interest in the history of the South" item and the "Civil War doesn't mean much to me personally" item correlated at only about 0.20 with the Civil War knowledge item, which suggests that Civil War knowledge is not a good proxy for sentiment about the Civil War or Southern history. Replication code is here, with a URL for the data: https://www.ljzigerell.com/wp-content/uploads/2017/07/Fall-1994-Southern-Focus-Poll.docx [UPDATE:] https://www.ljzigerell.com/wp-content/uploads/2017/07/Fall-1994-Southern-Focus-Poll-1.docx

    The "interest" item performed less well for the item about prohibiting private use of the Confederate flag, but not in a way that contradicts the above inferences: a positive coefficient and a p-value of p=0.188 with the racial attitude control, and a positive coefficient and a p-value of p=0.233 without the racial attitude control, with p-values a bit smaller (0.110 and 0.148) if participants from Kentucky and Oklahoma are excluded. [UPDATE:] I had incorrectly used logit instead of linear regression to predict the outcome variables for the Fall 1994 Southern Focus Poll. The p-values for the "interest" items using linear regression are less than 0.05 for the use of the Confederate flag items. Nothing else reported above for that poll changed.

  3. Comments on Strother et al. (2017) and Responses to Strother’s criticisms of Wright & Esses (2017):

    I begin by noting clear inadequacies in the Strother data for answering the proposed questions. Then I respond to a number of statements (see here) by Strother regarding our paper, which can be found here.

    Strother et al. (2017) analyses three representative datasets in an attempt to answer why people support the Confederate battle flag. The first sample is from 2004. This is notable because attitudes toward the Confederate battle flag are not stable and clear evidence of declining support for the flag have been presented. See here from August 2017 relative to here from June, 2015. Notable events have happened within even this short time frame (e.g., the A.M.E. Church shooting). Since Strother’s interest is in contemporary attitudes, a dataset 14 years old simply isn’t adequate.

    The general question is whether pride in the South (i.e., Southern pride) or negative racial attitudes toward Blacks are better predictors of support for the Confederate battle flag. The 2004 dataset from Georgia contains 341 representative Whites. Strother incorporates a measure of racial resentment, two single item measures of racism, a historical knowledge quiz consisting of whether a participants claims to know who William Sherman is and whether a participant can name zero, one or more than one civil war battle. There are a plethora of control variables included with no explanation for why they should or were included.

    The racial resentment scale seems reasonable enough, although Strother never provides any psychometric information about the measures used. We are left to assume that they are internally reliable, and that they are valid measures of what Strother intends to measure. Notably absent is one of the racial resentment items, which Strother noted was excluded, but he provides no explanation for why this item was excluded.

    Strother claims that the two historical knowledge items are a measure of Southern pride, but this assumes a very narrow definition of pride. Given the role of the Confederate battle flag is a pop-culture icon in food, music, TV’s shows, and movies, it is not rational to tie Southern pride to a naïve and insidious definition of it, which Southerners would largely disagree with. We diverge from Strother in that we see Southern pride as numerous qualities ranging from pride in ones cultural foods, pride in the values embedded within the South (e.g., work ethic, traditionalism, honor), pride in Southern rock and country music, pride in the “good ol’ boy” ethos displayed in The Dukes of Hazzard, and pride in the stereotype of the “Southern gentleman”. Further, Strother provides no evidence that his two knowledge items are a valid representation of any underlying latent construct, much less pride. Anecdotally, I asked a few Canadians to name some civil war battles, which they were easily able to do. I informed them, that according to Strother, they had strong Southern pride.

    There are some unexplained elements to this measure. First, there is no explanation for how these items, measured on distinct scales, were combined into one measure. Second, Strother artificially created a ceiling effect by categorizing two or more named civil war battles into one category, which would better have been left as a continuous variables. Third, we have no way of knowing whether respondents simply claimed to know William Sherman without actually knowing who he was. The survey was conducted via phone where socially desirable responding would be appealing (i.e., not wanting to appear unknowledgeable).

    The racism items are notably divided up into three different variables. Three are incorporated in the measure of “racial resentment”, opposition to interracial marriage takes on its own variable, and denial of racial disadvantage takes on its own variable. No explanation is given for why this analytical choice is made. I imagine that a factor analysis would confirm that these 5 items are all measuring a single underlying construct and the VIF’s of the three-item variable and the two single item variables are likely quite high. Seemingly arbitrary control variables are included. For example, no explanation is given for why owning your own home was important to control for in the analysis, the past literature never indicated any rationale for this. Artfully selecting control variables can be a method of p-hacking.

    The DV in this analysis was whether a respondent preferred the 1956 flag, the 2003 flag, or the newly designed 2004 flag. Notably, the 2004 flag is the national flag of the Confederacy and this category was combined with the 2003 flag to make a single category for comparison to the 1956 flag. No explanation was given for why a multinomial logistic regression would not be preferred over artificially combining categories.

    The second dataset is from 2014 and contains 581 White South Carolinians. Again, there is no measure of Southern pride and now the racial resentment measure is a single item. The control variables are different and no explanation is given to why. Additionally, other controls, such as gender, have never been associated with flag support, and it is unclear why this unassociated variable should continually be included analyses. Including non-relevant control variables can alter the parameter estimates and the statistical significance of findings. It is akin to including “junk” in the model.

    The ANES 2008 survey has no measure of support for the Confederate flag so I do not comment on it as it is irrelevant for answering the associated question. In general, I find the approach in Strother et al. (2017) to be no better in terms of measurement or conceptualization than the list of previous studies that have always excluded any measure of Southern pride from their work (e.g., Clark, 1997; Reingold & Wike, 1998; Orey, 2004; Cooper & Knotts, 2006). Wright & Esses (2017) is the only study to incorporate Southern pride, multi-items measures of constructs, and to incorporate controls for socially desirable responding.

    Below I respond to a number of claims made by Strother. These claims can be found here within an exchange between Strother & Zigerell (here).

    1. Strother claims that our paper suggests that negative racial attitudes don’t explain White support for the Confederate battle flag.

    It is true that we claim that racial attitudes cannot fully explain support for the flag but we never claim that racial attitudes do not partially explain support for the flag. Our paper explicitly states, “Among the White sample, positive attitudes toward the Confederate battle flag were significantly related to more political conservatism, more Southern pride, more negative racial attitudes toward Blacks, and less motivation to control prejudice.” Our model accounts for approximately 50% of the variance in White support for the flag.

    2. Strother claims that our paper is based upon an observational analysis of a convenience sample.

    Our sample consists of 417 White Southerners from the states of Georgia, Mississippi, and South Carolina. The sample was recruited through the Mechanical Turk platform. Mturk samples are very diverse (Buhrmester, M., Kwang, T., & Gosling, S. D., 2011), valid for political research (Clifford, S., Jewell, R. M., & Waggoner, P. D., 2015), and similar to national samples in various demographic characteristics (Huff, C., & Tingley, D., 2015). It is true that Mturk samples tend to be younger, more liberal, and more educated than the general population (Paolacci, G., & Chandler, J., 2014). This concern is explicitly addressed in the discussion section of our article.
    In contrast to national samples with pre-selected variables, our sample does provide the benefit of being able to include reliable and valid multi-item measures of constructs rather than single item measures that are typical of representative datasets.

    3. Strother claims that our measure of racial attitudes is inadequate.

    Our measure of blatant racial attitudes toward Blacks is similar to other blatant measures used in the literature (see Brigham, 1993; McConahay, Hardee, & Batts, 1981). It contains 7-items with good internal consistency and high factor loadings on the latent construct. We intentionally used a blatant measure rather than a “symbolic” or “modern” measure specifically because we are interested in racism, not aspects of the conservative ideology such as work ethic or merit-based systems. There are several empirical, logical, and methodological problems with symbolic racism as a construct (see Sniderman & Tetlock, 1986) that exist precisely because researchers imbed political values into these scales (Chambers & Schlenker, 2015). Chambers et al. (2013) manipulated race (black or white) and ideological position (liberal or conservative) of targets and found that modern racism was associated with negative attitudes toward liberals (whether white or black) and positive attitudes toward conservatives (both white and black)—a good indication that modern racism measures aspects of political ideology rather than racism. Relative to the reliance on single item measures prevalent in this literature, our scale is a considerable improvement and the literature suggested that blatant prejudice was the best predictor of confederate battle flag support relative to symbolic prejudice (Orey, 2004).

    4. Strother claims that controlling for blatant racism does not mean that the variance is accounted for by southern pride is “innocuous”.

    It is the case that we have removed variance accounted for by blatant racism. See our note above regarding “symbolic” or “modern” racism measures. To truly understand whether Southern pride is innocuous or not (and this probably is person dependent) we would need to evaluate various underlying motives for Southern pride. Many cultural phenomena (e.g., country music) have been associated with the South that could potentially be imbedded within Southern pride and the symbol has been a pop-culture icon since the 70’s.

    5. Strother later claims our paper shows a relationship between negative racial attitudes and support for the Confederate battle flag.

    Yes, it does. This is explicit in the manuscript.

    6. Strother suggests that his paper has representative state samples with good measures and consistently show a strong relationship between prejudice and support for the flag. A strong point for political science as a discipline is its use of representative samples, but state samples can’t be considered representative if your claims are made about the South in general, which is comprised of 11 in the traditional conceptualization or 13 in modern sociological works (see Knotts & Cooper, 2010). Second, two datasets in Strother et al. are over a decade old. If we are interested in contemporary attitudes toward the flag, shouldn’t we use data from today, rather than a generation ago? There are other examples of inadequacies in Strother’s data. For example, zero of the datasets used had measures of Southern pride and the 2008 ANES survey had no measures of attitudes toward the flag. Strother makes numerous claims about Southern pride throughout and interprets the ANES survey as if “warmth toward Southerners” were somehow a proxy for attitudes toward the flag.

    7. Strother claims that some southern heritage variables go in the wrong direction. Specifically, he claims that knowledge of civil war history is negatively correlated with positive attitudes toward the flag.

    This may very well be true, I don’t know because it has yet to be tested. Strother selected two “knowledge” questions out of multiple available in the Georgia dataset. One asked participants to name a civil war battle. If a participant named zero, zero points, if they named one, one point, if they named two, two points. This item has no evidence of reliability or validity as a measure of civil war knowledge. The second question asked participants if they knew who William Tecumseh Sherman was, but there is no check to ensure that those saying yes were not just responding in a socially desirable way and, in fact, have no idea who Sherman was. Even worse, Strother interprets the items as a measure of Southern pride, which does not encapsulate any conceptualization of pride or even identity that I ever seen. See Reed’s book “Southerners” for his conceptualization or the vast literatures on patriotism or social identity for others. We don’t know why Strother selected these questions from the available knowledge items rather than creating an aggregate of all the knowledge items. Worse still, the analytic decision to create an artificial ceiling of scores by not counting any named battle beyond two creates a restricted range and it is possible that flag supporters could have been overrepresented in the range of scores ignored by Strother. He further suggests that his measure captures the “latent concept of pride” but if that is the case, why not analyze the data using a latent variable model so that readers can see how well the items tap into the same latent construct, even if it isn’t necessarily Southern pride?

    8. Strother claims that warmth toward Southerners is associated with racial prejudice.

    I imagine we could consider warmth toward Southerners to be a form of in ingroup identification. If so, there is a large literature linking stronger ingroup identification with outgroup prejudice. This shouldn’t be specific to racial prejudice and Reed (2008) has shown that Southerners harbor more negative attitudes toward other outgroups such as Northerners. Strother’s finding is also likely a result of Southerners displaying a traditional value orientation and attitudes toward outgroups reflect attitudes toward value-violating groups. This would align with Chambers et al. (2013) described earlier.

    9. Strother claims that the majority of Southern Whites harbor negative racial attitudes towards Blacks. We didn’t find this in our paper. Piston (2010), cited by Strother as evidence for this claim also doesn’t confirm this. Rather, Strother chooses to interpret positive attitudes as negative solely because they are relatively more negative than attitudes toward an alternative group. There is also a clear publication bias in the racial discrimination literature and many non-published studies fail to support Strother’s claim that the majority of Whites harbor anti-Black attitudes (Zigarell, 2018).

    10. Strother argues that Southern pride is not innocuous but is a proxy for racial prejudice.

    It is associated with racial prejudice in the aggregate, and this can clearly be seen in our paper. However, we diverge from Strother in that we see Southern pride as potentially consisting of numerous qualities ranging from pride in ones cultural foods, pride in the values embedded within the South (e.g., work ethic, traditionalism, honor), pride in Southern rock and country music, pride in the “good ol’ boy” ethos displayed in The Dukes of Hazzard, and pride in the stereotype of the “Southern gentleman”. In contrast, Strother isolates Southern pride into one thing: pride in the Confederacy and slavery. If we wish to separate our innocuous pride from pride in the Confederacy specifically, we will need to create distinct measures for these constructs, which currently do not exist. Post-hoc selecting of arbitrary items from national datasets is not a viable alternative to basic psychometric work. Strother further claims that because warmth toward Southerners is associated with prejudice toward Blacks, this undercuts the claim that Southern pride is innocuous. We don’t see how. Without any measure of Southern pride, this conclusion cannot be drawn. Interestingly, we also find that the zero-order correlation between Southern pride and positive attitudes toward the Confederate battle flag is positive among Blacks (albeit small: r = .2). This could be an artifact of the small sample of Blacks or it could be preliminary evidence that pride in some other aspects of Southern culture (unrelated to the Confederacy) is associated with the Confederate battle flag. This would be reasonable given the flag’s pop-culture status for almost half a decade.

    11. Strother reiterates his claim here, “I’ll say again that racial prejudice goes much further in explaining white support for the Confederate flag than does southern pride”, but again, Strother has never measured Southern pride so this is just conjecture.

    *All comments are my own and do not necessarily reflect those of the second author.

    -Joshua D. Wright

  4. Small error: "flag’s pop-culture status for almost half a decade" should reflect "flag’s pop-culture status for almost half a century".

    • Thanks for the great comment, Joshua.

      I concur with a lot of what you wrote, but I don't think that Strother et al. had control over the decision to not code the Civil War battles item continuously, because the item as reported on in their appendix wasn't coded continuously (see below). Of course, as you indicated, the item isn't very useful as a measure of Southern pride. There's also the issue of whether responses like "Battle of Bunker Hill" were coded as "NAMES OTHER BATTLE", although such miscoding in terms of correctness might not necessarily bias inferences about associations with flag support.

      ---

      Do you know the names of any Civil War battles?
      [INTERVIEWER: DO NOT READ CATEGORIES]
      1. NO
      2. YES, GETTYSBURG
      3. YES, NAMES OTHER BATTLE
      4. YES, NAMES MORE THAN ONE BATTLE
      5. YES, BUT WON'T GIVE NAME

Leave a Reply to Joshua D. Wright Cancel reply

Your email address will not be published. Required fields are marked *

*

This site uses Akismet to reduce spam. Learn how your comment data is processed.