By: Sean Stevens, HxA Research Director and Jonathan Haidt, HxA Board Chair
Abstract: A series of essays published in March argued that American college students are just as supportive of free speech as they have ever been, so there is no “free speech crisis” on American college campuses. In this essay we show that this optimistic reading of the most widely used longitudinal dataset (the GSS) greatly overestimates political tolerance because of a methodological artifact: it asks about speakers that are not very controversial anymore, such as a “communist” or a “homosexual.” When we examine data about speakers that are strongly disliked today, we find much higher rates of political intolerance, particularly among college students on the left. We argue that this political asymmetry tells us nothing about the left or right in general, but rather about who is feeling psychologically insecure in the last few years. We think these re-interpretations of the GSS, along with an analysis of two recent datasets, undercuts the skeptics’ main argument and contradicts their thesis, while pointing to productive new ways to address the conflicts roiling many campuses.
Is there a “free speech crisis” on American college campuses? A half-dozen essays were published recently disputing the claim. (See our list of those essays, and our response to it, here. See Sachs’ response to our response here.) The wave of essays was launched by an analysis by Justin Murphy of items on the well-respected General Social Survey (GSS). That survey includes questions that have been asked every two years since 1972 about American attitudes toward controversial speakers. The GSS data clearly shows a steady increase in tolerance, decade after decade, as you can see in the graph below, which is a graph produced by the GSS site, as shown in Jeffrey A. Sachs’ essay titled “The ‘campus free speech crisis’ is a myth. Here are the facts.”
Figure 1. A graph of GSS data, reprinted in Sachs’ essay titled “The ‘campus free speech crisis’ is a myth. Here are the facts.”
Here is Sachs on the implications of this graph (and other similar ones):
On almost every question, young people aged 18 to 34 are the most likely to support free speech. Check out the data for yourself. Not only are young people the most likely to express tolerance for offensive speech, but with almost every question posed by the GSS, each generation of young people has been more tolerant than the last.
Judging by the GSS data, it sure looks like there’s no problem with young people today; in fact, the skeptics say, those who are alarmed about campus trends have things exactly backwards. Matt Yglesias said this in his essay at Vox titled “Everything we think about the political correctness debate is wrong.” His subtitle asserted that “Support for free speech is rising, and is higher among liberals and college graduates.”
But there is a major problem with the way the GSS assesses political tolerance: It uses what is known as the “Stouffer method” (reviewed in more detail below)––a set of questions that asks respondents about whether a variety of groups, preselected by the researcher, should be allowed to speak in a community, teach at a college or university, or have a book in the local library. Samuel Stouffer first developed this method in the 1950s to assess attitudes about communists, socialists, and atheists––groups specifically chosen because they were controversial and non-conformist at that time.
Starting in 1972 the GSS included Stouffer’s original 3 groups, asking about the following targets:
- Somebody who is against all churches and religion. (An atheist)
- A person who favored government ownership of all the railroads and all big industries. (A socialist)
- A man who admits he is a communist.
In the 1973 survey, “a man who admits that he is a homosexual” was added as a target. In 1976 two more targets were added: “a person who believes that Blacks are genetically inferior” and “a person who advocates doing away with elections and letting the military run the country.” Finally, in 2008 “a Muslim clergyman who preaches hatred of the United States” was added as a target. The skeptics are correct that the GSS data shows tolerance of these targets, averaged together is rising, decade after decade, but does that show a general rise in tolerance for controversial speakers? Or might it merely show that the specific targets people find upsetting in one decade often become less upsetting in later decades, to be replaced by new targets?
That was precisely the question asked by Sullivan, Piereson, and Marcus (1979) who made the simple conceptual point that you can’t measure tolerance unless you know for sure that the people you are surveying oppose the group you are asking about––which the Stouffer method does not do. Sullivan et al. defined tolerance as “a willingness to permit the expression of those ideas or interests that one opposes.” They added that tolerance presumes opposition. To allow expression of doctrines with which one agrees, or to which one is indifferent, does not manifest political tolerance. “Therefore any attempt to measure tolerance must first determine the presence of opposition” (Sullivan, Piereson, and Marcus, 1982, p. 23, emphasis added; see also, Gibson, 2006; Sachs does acknowledge this point in his essay). When Sullivan et al. allowed people to choose from a list of groups the ones that they opposed, and then looked at their levels of political tolerance, they found there had been little increase in tolerance since the 1950s.
In this essay we argue that the insights of Sullivan, et al. (1979; see also Sullivan et al., 1982) are crucial for understanding events on campus today. Once those insights are incorporated, the skeptics’ strongest arguments become considerably weaker, and in some cases their implications reverse. We show––as we have in our previous posts––that there is evidence that college students today still value free speech in the abstract, but they are willing to deny it to members of groups they dislike or consider controversial. In other words, something is changing on campus, and the skeptics’ counter-arguments are based in large part on the unreliable Stouffer approach, and on survey questions that ask about free speech in the abstract, without pitting it against other values.
MEASURING POLITICAL TOLERANCE: THE STOUFFER METHOD
The GSS employs the Stouffer (1955) method of assessing political tolerance, where tolerance is studied in relation to particular groups selected by the researcher beforehand. Stouffer’s original study, conducted in 1954, focused on attitudes towards communists, socialists, and atheists, all of whom at the time were considered “non-conformists” in American society. Substantial majorities were unwilling to grant these groups full civil liberties, and Stouffer concluded that perceived threat from the group in question was one of the more important factors in explaining an individual’s willingness to restrict, and possibly take away, the rights of the threatening group.
The application of the Stouffer method for assessing political tolerance in the GSS is not limited to atheists (or “anti-religionists”) and communists. By 1976 the political tolerance of homosexuals, militarists, and racists was also assessed. Subsequent work analyzing the responses to these items on the GSS (see e.g., Cutler & Kaufman, 1975; Davis, 1975; Erskine & Seigel, 1975; Nunn, Crockett, & Williams, 1978) has reported broad overall increases in political tolerance among the American electorate (for more recent findings see e.g.: Nie, Junn, & Stehlik-Barry, 1996; Schafer & Shaw, 2009; Schwadel & Garneau, 2014). Since 1976, tolerance of homosexuals has increased the most, while tolerance of racists has increased the least. These increases in tolerance were positively correlated with more education and greater endorsement of individualistic values, and negatively correlated with changes in empathy (Twenge, Carter, & Campbell, 2015; see also Schwadel & Garneau, 2014). In 2008, the GSS added a sixth target: a Muslim clergy member preaching hatred of the USA.
Yet, the claim that tolerance in the United States has gradually, and consistently, increased over time has long been disputed (e.g., Gibson, 1992; Mondak & Sanders, 2003, 2005; Sullivan, Piereson, & Marcus, 1979). Almost 40 years ago, Sullivan et al. (1979) argued that the broad increases were illusory and mainly the result of Americans expressing less dislike for atheists, communists, and socialists in the 1970s. When Sullivan et al. focused their assessments of political tolerance on groups that were known to be disliked by the respondents, a method described in more detail below, they found that large majorities of Americans were willing to restrict the political rights of groups they disliked.
As Sullivan et al. (1982) described (p. 6), studies employing the Stouffer method…
…do not assess tolerance as a general attitude or disposition to permit the expression of opinions or interests that one opposes. In studies of attitudes toward, and tolerance of, a particular group (such as communists), it is unlikely that all respondents oppose the group equally. Thus one precondition that our more inclusive definition of tolerance requires — opposition toward the group in question — will exist for some respondents and not others. Therefore, the conclusions of such studies bear more directly on the group than on tolerance in a more universal sense.
Gibson (1992) reached a similar conclusion, arguing that the Stouffer method is not a valid measure of tolerance in general because it does not assess the degree to which someone is putting up with a disliked group. Gibson also contended that while it was appropriate to draw cross-sectional conclusions from research assessing how responses to the Stouffer items of the GSS have changed over time (e.g., comparisons across different cohorts at one point in time), it is not appropriate to draw longitudinal conclusions about changes in tolerance in the United States in general (e.g., conclusions about changes in levels of tolerance over time; see also Mondak & Sanders, 2003). Yet that is exactly what some of the skeptics do with the GSS (see here).
MEASURING POLITICAL TOLERANCE: THE LEAST-LIKED GROUP METHOD
One question that quickly emerged following Stouffer’s (1955) original study was: how is American democracy sustained, if most people are somewhat politically intolerant? Shortly after Stouffer’s (1955) analyses, Prothro and Grigg (1960) and McClosky (1964) independently reported that Americans showed a high degree of agreement when asked about the general principles of democracy (e.g., “Every citizen should have an equal chance to influence government policy” (Prothro & Grigg, 1960, p. 282)), but that this consensus disappeared when more context and specifics were added (e.g., asking if “In a city referendum, on tax-supported undertakings, only taxpayers should be allowed to vote” (Prothro & Grigg, 1960, p. 282); for other examples, see Lawrence, 1976).
Such findings suggested that the Stouffer method was only measuring tolerance of specific groups, and not tolerance in general. As a result, Sullivan, Pierson, and Marcus developed what has come to be called the least-liked group method (see Sullivan et al., 1979; 1982). With this method, researchers present a list of groups that are likely to be unpopular, either to the political left, the right, or both sides, and then they ask respondents to choose their least-liked group [for a variation on the least-liked groups method, see footnote 1]. Respondents are allowed to provide their own least-liked group via an open-ended response if they do not find it on the list they are presented with. The least-liked group method makes it far more likely that one is assessing a general attitude or disposition toward political tolerance. Gibson (2006, p. 102) has referred to it as “the standard technology” for measuring tolerance of groups people find objectionable [For concerns about the least-liked group method, see footnote 2].
Research employing the least-liked group method typically finds that:
- Large majorities are politically intolerant of their least-liked political groups (e.g., Gibson, 2006, 2008, 2013; Sullivan et al., 1982);
- The least-liked group change in response to political events (e.g., Gibson & Gouws, 2001); and
- The selection of targets for political intolerance is primarily driven by ideological identification (e.g., Gibson, 2006; Sullivan et al., 1982). People on the right are more politically intolerant of groups on the left, while people on the left are more politically intolerant of groups on the right (see e.g., Gibson, 2006; 2008; 2013; Sullivan et al., 1979; 1982).
Sullivan et al. (1982) also concluded that (p. 249-250):
It is generally true that the sociological variables (age, religion, race, class, sex, and education) are more important in explaining the target groups picked, while certain political and psychological variables (e.g., perceptions of threat, dogmatism, self-esteem, and ideology) are more helpful in understanding variations in levels of tolerance.
Now, finally, with this long introduction done, we can turn to two recent surveys of political tolerance expressed by current college students.
TOLERANCE OF SPEAKERS IN THE FIRE/YOUGOV DATASET
The FIRE/YouGov 2017 survey data (N = 1,250 current undergraduate students) we previously referenced in Part 2 contained a hybrid version of the Stouffer method and the least-liked groups method. Respondents were asked: “I may want my college or university to tell a guest speaker he/she is no longer invited to speak if the speaker is: (check all that apply)” and were then presented with a list of 64 specific targets. Although the list of speakers was preselected by the researchers and the presence of opposition was not assessed in advance, the long list made it likely that most respondents would find at least one target they would consider disinviting (see Gibson, 1992).
The left-hand panel of Figure 2 below depicts the 6 speakers that respondents who had self-identified as “very liberal” would consider disinviting (We chose these 6 speakers because they elicited the highest percentages of “yes” responses when participants were asked if the speaker should be disinvited.) The right-hand panel does the same for very conservative respondents. One interesting caveat is that three speakers were in the top six for both left and right: sexist, racist, and Holocaust denier. We therefore removed those three from consideration [see footnote 3 for the mean percentages of all respondents, by self-identified ideology, to these potential speakers] and replaced them with the next three speakers that very liberal respondents reported they would consider disinviting. This produced two lists of six that were unique for each side.
Figure 2. FIRE/YouGov 2017 data on support for the formal disinvitation of speakers from a college or university. The left panel shows the 6 potential speakers who were most offensive to students on the left; the right panel shows the 6 potential speakers who were most offensive to students on the right.
The difference between the panels is stark. For the speakers highly disliked by the left, 40% or more of “very liberal” college students would consider disinviting them, and in some cases this was over 60% (the average was 55%). For the speakers highly disliked by the right, the percentages of “very conservative” students considering a disinvitation are much lower (average of 21%). Figure 2 is inconsistent with the claims by the skeptics that current college students are extremely tolerant, and that students on the left are particularly noteworthy for their tolerance. Those claims are consistent with the Stouffer method, which asks mostly about speakers that are more disliked on the right, but they are not consistent with findings using a method that is more likely to provide respondents the ability to identify a disliked target.
Speakers who are regarded as homophobic, transphobic, or Islamophobic can be thought of as the communists or atheists of today. They are the most intensely disliked, particularly, it seems, on college campuses. If we want to know whether college students are more tolerant today than they were in generations past, it makes little sense to ask them if they’d tolerate a communist, an atheist, or a homosexual making a speech. The rising lines of tolerance in Figure 1, using the Stouffer method, do not show that young people are becoming more tolerant of speech across the board. Rather, an approximation of the least-liked groups method suggests that students on the left are intolerant toward many speakers, particularly those who, presumably, are seen to violate their core values (see e.g., Graham & Haidt, 2012). Most of the groups that current students––particularly liberal students––would consider disinviting are linked to bigotry, but notice that the “anti-feminist” is not anti-woman; the speaker is opposed to feminism, a family of political/intellectual movements. Nearly half of the very liberal students surveyed said that such a speaker should not be allowed to offer such critiques on campus.
One might argue that the speakers opposed by the left are objectively worse than the speakers opposed by the right. It is likely that students on both sides believe–and can explain–why the speakers they dislike are uniquely awful and dangerous, but suppose we accept the claim that the FIRE/YouGov survey happened to pick a worse rogue’s gallery for the left than for the right, so the difference between the panels is an artifact of the survey design (as were many of the results obtained using the GSS employment of the Stouffer method). In order to address this concern, we chose five pairs of speakers that were conceptually equivalent, coming from the same category but differing in their political polarity:
- Sarah Palin and Barack Obama (both are specific politicians who have high name recognition and were on the ballot in the 2008 Presidential election).
- A pro-police/Blue Lives Matter activist and a Black Lives Matter activist (both are political activists on opposing sides of the same, current culture war issue).
- A gun rights activist/NRA member and a gun control activist (both are political activists on opposing sides of the same, current culture war issue; Note: this data was collected in the Spring of 2017, before the school shooting in Parkland, Florida).
- A Fox News reporter or show host and an MSNBC reporter or show host (both are cable news reporters or pundits on the right and left respectively).
- A Conservative Republican and a Liberal Democrat (both are party affiliates on the right and left and respectively).
Figure 3. FIRE/YouGov 2017 data on support for the formal disinvitation of speaker from a college or university. Limited to speakers that can be placed into conceptually equivalent pairs.
When we make this correction, the difference between the panels shrinks considerably. In fact, the first two pairs (Palin/Obama, and Blue-lives/Black-lives) are mirror images of each other. However, on the other three speakers, we see the very liberal students showing much higher levels of intolerance than the very conservative students. The cleanest possible comparison is that 18% of very liberal students would consider disinviting a speaker merely for being a “conservative republican”, whereas only 8% of very conservative students would consider disinviting a speaker merely for being a “liberal democrat.” Once again, this finding contradicts the skeptics’ claims that college students on the left are more tolerant or more supportive of free speech than are college students on the right.
TOLERANCE IN THE CATO/YOUGOV DATASET
Like the FIRE/YouGov survey, the Cato/YouGov 2017 survey on Free Speech and Tolerance (Total N = 2,300, including an oversample of 769 current college students and college graduates) also asked about whether certain speakers should be allowed to speak on campus. Although the Cato/YouGov survey employed the Stouffer method to assess whether certain types of speakers should be allowed to speak at a college or university, the speakers they preselected help demonstrate one of our broader points: that levels of political tolerance depend on how political tolerance is measured. Thus, in contrast to the FIRE/YouGov survey, 13 speakers were described, and, in our estimation, 9 of the 13 speakers offered were more likely to offend the left. One of the speakers (“advocates violent protest”) was likely to offend both sides. We therefore picked the three speakers who were more likely to offend the right, and paired them with three of the speakers offensive to the left.
Figure 4. Cato/YouGov Free Speech and Tolerance 2017 data; Percent of current college students who say that a speaker should not be “allowed to speak” at their college.
Once again, we see that students on the left are more likely than students on the right to want their school to disinvite the speakers that their side dislikes. In fact, the support for disinviting controversial speakers was so high and consistent on the left that “very liberal” students were more likely than “very conservative” students to want to disinvite twelve of the thirteen speakers — even two of the speakers we think are more likely to offend the right (as you can see in the right-hand panel of figure 4). These findings are incompatible with the skeptics claim that today’s college students are highly tolerant of speech they disagree with, and the claim that the left is more tolerant than the right [see footnote 4].
PUTTING IT ALL TOGETHER: WHAT IS CHANGING AND WHY?
The skeptics tell a story of high and uniformly rising tolerance on college campuses. They offer a narrative in which all is well for free speech on college campuses, and that those who raise alarms are engaging in a “moral panic”––an over-the top alarmist frenzy, fueled by right-wing media. We agree that the right-wing media is now doing what it can to raise alarm, and that it often exaggerates or distorts events on campus, but that does not mean that all is well on campus.
We think the work of Sullivan, Piereson and Marcus (1979, 1982; see also, Gibson, 2006, 2008, 2013; Gibson & Gouws, 2001; Mondak & Sanders, 2003; Sullivan & Transue, 1999), gives us a more accurate picture of what factors impact political tolerance, and of why the GSS data is not the optimal method to assess such factors.
In fact, Sullivan et al. give us far more than a reason to be skeptical of the skeptics. They actually give us a simple yet powerful multivariate model that may be the key to understanding why the dynamics on many campuses have changed recently (as we argued in our previous post).
Based their review of the scholarly literature and on maximum likelihood confirmatory factor analysis (a form of structural equation modeling, see e.g., Gunzler & Morris, 2015) of a data obtained from a national sample, Sullivan et al. (1982) proposed the model in figure 5. This model has stood the test of the time, although there remain some “enigmas of intolerance” [see footnote 5].
Figure 5. Sullivan et al. (1982) model of political tolerance. There are three pathways by which political tolerance can go up or down in a particular individual.
The model shows three separate inputs, three factors that can raise or lower political tolerance:
- Psychological Insecurity: The central horizontal line in the model says that people who are feeling psychologically secure are more likely to be tolerant of other people’s speech, whereas people who feel anxious, insecure, and vulnerable are more likely to be intolerant and want protection from speakers.
- Support for democratic institutions and procedures: This is a moderating variable, impacted by an individual’s level of psychological security. People who strongly support democratic institutions and procedures can remain tolerant even if they feel psychologically insecure. But people who do not strongly support such institutions and processes will become more intolerant as their feelings of insecurity increase. They don’t trust the system to work properly.
- Perceptions of group threat: If the speaker under consideration is judged to pose a direct threat to oneself, one’s group, or one’s personal values, then people will be much less likely to tolerate public speech by that speaker. Additionally, if the speaker is perceived as someone who violates norms and expectations about proper behavior and procedures (or if the speaker is a member of a group that is perceived in this way), then many people, including those who typically politically tolerant, may react with political intolerance.
Putting this all together, we can now understand why it might be the case that the dynamic on some college campuses has changed rapidly, just since 2013:
- Psychological insecurity. Rates of all violent crimes are way down since the early 1990s, on campus and nationally. Yet perceptions of physical threat and danger may be up. Rates of anxiety disorders and depression have increased dramatically among adolescents in general, and college students in particular, particularly female college students. Furthermore, the series of videos of police shooting unarmed black men, which began circulating widely in the years after the (unfilmed) Michael Brown killing of 2014, understandably increased psychological insecurity among black students.
- Support for democratic institutions and procedures. This moderating variable has been declining in democracies all over the world, as shown in the work of Roberto Stefan Foa and Yascha Mounk (see also here, here and here; for criticisms of Foa and Mounk’s analyses see here). It may have also taken a tumble for young Americans, particularly those on the left, during and after the US presidential election of 2016.
- Perceptions of group threat. Anyone who can in any way be linked to a student’s insecurities may seem more threatening. Students in recent years have made heavy use of concept creep to link disliked speakers to the worst possible sins and threats. For example, students at Pomona college called Heather Mac Donald “a fascist, a white supremacist, a warhawk, a transphobe, a queerphobe, [and] a classist,” without offering evidence for these charges, in order to justify shouting her down. Furthermore, the increasing sophistication and aggressiveness of right-wing trolls, empowered by the internet, has led to a real increase in threats, provocations, and harassment, directed mostly from off-campus sources to students and professors on campus.
To conclude: the skeptics case is based heavily (though not entirely) on data showing a steady increase in political tolerance toward controversial speakers. However, as we have shown, this increase relies on using the Stouffer method of tracking attitudes toward fixed groups, such as communists and homosexuals, who just don’t elicit the objections that they did many decades ago. When we apply a close approximation of the least-liked groups method, on data collected in 2017 (and thus after the key 2015 transitional year), we find very high levels of intolerance toward speakers that are offensive to college students on the left. We do not have longitudinal data from before 2015 to show that these high numbers are an increase from previous years, but these high numbers contradict the skeptics claims that political tolerance among today’s college students is high, increasing, and higher on the left than the right.
More importantly, we have drawn on political science work from the 1980s to offer an account of exactly why political intolerance may be increasing on the left on college campuses in recent years: because more and more students are feeling psychologically insecure, losing faith in democratic institutions, and facing (or at least hearing about) real threats from off-campus right-wing sources. The best way to increase political tolerance on campus may therefore be to increase feelings of psychological security, physical security, and trust in local and democratic institutions.
1. A variation on the least-liked group method asks respondents to identify more than one least-liked group and rank them. Findings using this method typically show slight increases in tolerance as one goes from the least-liked group, to the second least-liked, etc.
2. It is important to note that, although the least-liked group has been declared the “standard technology” for assessing political tolerance (Gibson, 2006, p. 102), it is not immune from criticism (e.g., Nie et al., 1996; Wagner, 1986). As Petersen, Slothuus, Stubager, and Togeby (2011, p. 583) note:
First, the ‘least-liked’ group will often be very extreme and violent. Whether or not one rejects the rights of such extremist groups does not bear on the subject of political tolerance with the same weight as if one were to reject the rights of groups who themselves observe other groups’ rights. By way of example, it is more consequential how citizens react to Muslims in general than how they react to Islamic fundamentalists; but the ‘least-liked’ approach makes us focus on the extreme group. Hence, the use of the ‘least-liked’ approach fails to grasp what has been called the ‘breath of intolerance’. Secondly, the ‘least-liked’ measurement strategy bypasses a fundamental question: can tolerance be expected to be limitless? When members of a group engage in violent and/or non-democratic behaviour, thereby denying others their basic rights, it might reflect a tolerant position to seek to curtail the activities of such extreme groups. This argument has been made by pointedly by Sniderman et al. (1989), but the idea is not new. Liberal philosophers since Hobbes and Locke have justified the concept of civil liberties by invoking the notion of the social contract. From the perspective of the social contract, a group’s right to freedom is deeply linked to its own observation of the same rights in relation to other groups.
In a more recent review on political tolerance, van Doorn (2014, p. 912) contends that the both the Stouffer method and the least-liked group method may fall short in measuring tolerance, because both methods reveal tolerance of a specific group or groups, but neither can reveal tolerance towards specific acts such as hate speech. In other words, support for civil liberties, such as freedom of speech, may find itself at odds with other important values (see e.g., Peffley, Knigge, Hurwitz 2001), and much of the survey-based research on political tolerance has not adequately captured the trade-off between different values that may affect an individual’s political tolerance.
3. Mean percentages across self-identified ideology for considering a disinvitation of the following speakers.
|Ideological Self-Identification||A Holocaust denier||Racist||Sexist|
4. We also looked at the mean percentages across self-identified ideology for all speakers. The table below presents these means. When all targets are included in the assessment – something we think would violate the objection precondition required for political tolerance to be in question – very liberal respondents, compared to all other respondents, are the most likely to consider a disinvitation.
|Very Liberal||Liberal||Moderate||Conservative||Very Conservative|
5. See e.g., Gibson, 2006), such as understanding the complex antecedents of group threat (see Gibson, 2006 for a review), explaining independence of social and political tolerance (Gibson, 2004; Gibson & Howard, 2007), and, obtaining a better understanding of why the intolerant are more likely to act on their attitudes than the tolerant (e.g., Marcus, Sullivan, Theiss-Morse, & Wood, 1995).
Opinions expressed are those of the authors. Publication does not imply endorsement by Heterodox Academy or any of its members. We welcome your comments below. Feel free to challenge and disagree, but please try to model the sort of respectful and constructive criticism that makes viewpoint diversity most valuable. Comments that include obscenity or that sound like a tirade or screed are likely to be deleted.