Measuring Self-Esteem

Self-esteem (SE) scales are some of the most commonly used tools in psychology.  The predominant SE test is the Rosenberg Self-Esteem Inventory (RSI), a short written test of questions relating to judgments of the self that relies solely upon self-report data. This test has been in use without modification for over half a century.  To say that psychologists have not gained sufficient insight into SE during that time to warrant an update or improvement on this tool seems inconsistent with the principles of constructive research.  In some ways, it is hard to fathom why psychologists have not at least added another dimension to the RSI using the same approach the authors used to create the first version. Unfortunately, this test has become so endemic in the SE research area that improvements have been absent even when research could have benefited from the availability of a more in depth, less biased, and more accurate assessment than the RSI can provide.

Since SE is a major component of the self-concept, and since we are living through our individual self-concept all of the time, SE must influence all aspects of our lives in one way or another.  Therefore, it is difficult to argue against the idea that greater understanding of the SE phenomenon would not only give rise to better testing modalities, but also have great benefit to SE research principles overall.  For example, researchers face significant difficulty understanding the etiology of, and developing efficacious therapies for, psychopathologies such as anorexia nervosa (AN) that possess strong linkage to SE. It is this author’s contention that the difficulties researchers face is due in part to a failure to truly understand and assess human fundamentals like SE. As a result, much published research implicates chronic low SE as a comorbid factor in disorders like AN without considering that chronic low SE itself can account for most, if not all, of the underlying drivers and characteristics of AN.  In fact, SE consistently presents as a key factor in the expression of eating disorders (see the link to Self-Esteem Meta-analysis).

Many psychologists defend the continued use of the RSI by pointing out that it does measure characteristics of SE and is relatively consistent over time.  While these are important aspects of a valid SE scale, the shortcomings of the tool create only a superficial assessment of SE and therefore leave much room for misinterpretation. For example, a sound scientific approach to research dictates that the testing tool remove as much subject bias from psychological testing as possible.  Unfortunately, self-reporting Likert scale tests like the RSI are extremely vulnerable to subject bias and the development of a new or improved assessment inventory is both desirable and necessary.

Flaws of the RSI

Although this author finds the RSI to be both flawed and overly superficial, my criticism is not so much of the RSI itself as it is of its widespread continued use with little or no improvement for decades. The RSI was originally created by a team of sociologists in 1965 who were working with a large subject population and who required a quick, easily administered, unidimensional SE test with face validity in order to employ multivariate analysis approaches to SE research.  The authors accomplished this goal fairly well in that their premises were well thought out and arguments for validation were well made.  However, even the authors would have recognized that they were creating a relatively superficial tool that therefore lacked specificity and sensitivity to individual differences.

This lack of specificity is apparent when you consider the fact that the RSI does not differentiate between high SE and defensive high SE, also known as narcissism.  By definition. The word “defensive” is one that should correlate more tightly with low SE.  Therefore, defensive high SE is actually a contradiction in concepts. Psychologists describe the narcissistic individual as possessing an over-emphasized sense of self-importance, a characteristic that runs contrary to the more balanced approach we associate with high SE. It is the author’s opinion that narcissists fall within the low SE group, but have developed compensatory presentational tactics that make them present as high SE individuals.  Unfortunately, the RSI cannot distinguish true high SE from narcissism based upon the short directive questions, a serious inadequacy in any SE test.

Some psychologists have questioned whether explicit SE scales like the RSI actually measure self-presentation rather than SE.  Considering the fact that Western societies have an undeniable emphasis on presentation, this is a valid question.  One doesn’t need to be a psychologist to notice the level to which human beings are capable of believing in a false representation of self, or falsely interpreting their own and others’ actions colored by those beliefs. As such, it is logical to assume that presentation impacts RSI scores, even if such have not been routinely parceled out of the data, and that it is highly desirable to remove this confounding variable from the mix.

Another criticism of the RSI is the fact that it presents with heavy within-test bias, a problem shared with many self-report Likert style test.  For instance, the way a subject answers one question influences how they answer another, unless they are prepared to provide obviously conflicting information.  For example, it is totally inconsistent to choose “agree/agree very much” for the statement “I take a positive view of myself” and then to choose “agree/agree very much” to a subsequent statement of “All in all I am inclined to feel that I am a failure.” Likewise, a positive answer to the statement “On the whole I am satisfied with myself” is not consistent with a positive response to the statement “I wish I could have more respect for myself.”  Some psychologists would point out that it is possible to feel satisfied on the whole but still wish for more self-respect, but on face value, most subjects completing this inventory would be aware of the contradiction and the gross generality of the statements does not allow for assessment of the degree of regard or the nuances impacting the judgments.  This gives rise to ambiguity in how any given subject interprets the questions, and since the responses fall into two opposing categories, this will have some biasing effect on the overall results.  It is not difficult to conceive that identifying oneself with one side or the other in the first few questions naturally predisposes the subject to maintaining continuity for future responses, either consciously or subconsciously.  Therefore, within-test bias is a significant confounding variable in the RSI, especially since the objective of the test is readily apparent to the subject taking it.

Developing a New SE Test

In considering the development of a new SE test that accounts for the deficiencies of the RSI, we are faced with two fundamental questions – which expressions of SE are we going to measure (the RSI measures the obvious overt expressions of SE as perceived by the individual) and how are we going to measure them?

The question of what we need to measure is theoretically easy to answer.  Psychologists agree that low and high SE individuals process information related to the self differently.  This means that subjects on opposite ends of the spectrum have different cognitive criteria that are reflected in the way stimuli are interpreted, the types of behavioral choices that are acceptable, and in personal preferences across a wide variety of situations.  Even a concept as simple as to what a person will attend to in a given social interaction will reflect these different criteria in some form or another. 

The problem that we face here is a significant lack of essential data.  SE researchers understandably tend to gather data related to their specific point of focus, but what is needed is a significant increase in our basic understanding of SE, how it develops, its purpose, and how it perpetuates our self-view. This not only requires gathering more in depth and multidimensional data in the lab, but also doing extensive field work.  There is only so much that can be learned by asking subjects questions in a controlled setting where presentational factors are highly activated. The main limitation is that researchers gather data related to the expression of a phenomenon but never get to witness the subjective expression of that phenomenon for themselves.  If the researchers were able to spend a couple of days observing someone with low SE compared to someone with high SE, they would not only collect significant data related to subjective expressions of SE in general but would also be able to observe these expressions in a real life setting with more validity.

The question relating to how we measure SE evokes the concept of implicit vs explicit testing.  The most obvious problem with explicit self-report tests is subject bias.  Implicit tests are in general more desirable because they account for this variable.  Unfortunately, psychologists tend to view implicit tests with skepticism because they do not correlate strongly enough with explicit tests like the RSI. This could easily be the case of a better approach being rejected because it doesn’t correlate strongly enough with a worse approach.  In addition, variance in implicit testing across individuals can impact the reproducibility required for much organizational SE testing, leading us to a false narrative that “SE is what the RSI measures” similarly to what has been found true for “Intelligence is what IQ tests measure.”  Despite the complexities of designing and validating implicit SE tests, any approach that helps address presentational and subject bias is preferable to an approach that doesn’t.

Both bias and the ease of administering the test would be addressed if an implicit paper and pencil test were developed.  The first requirement would be to have neutral, non-evocative questions as the research vehicle.  By neutral, I mean that the questions themselves do not give any specific indication of what is being measured. The questions in the RSI leave the subject in little doubt as to what is being assessed. By non-evocative, I mean that the questions are less likely to evoke subject bias by occluding the specific goal or purpose of the question.  For instance, asking a subject the question “I feel I am equal to others” requires them to make a personal assessment of their own self-worth. This opens the door for all manner of individual and societal bias and confounding variables to be evoked.  Identifying characteristics of individuals with low and high SE on other dimensions would reduce this interference.  For instance, if low SE people tend to prefer activities that can be done alone and those with high SE prefer group activities (this is totally hypothetical and for illustration purposes only), then scenarios could be devised to tap into these differentiating characteristics and thus mask or avoid perception of the test as centering on SE.

Because SE is strongly related to our perceived status in relation to the external world (and specifically other people), it follows that factors related to SE will be activated when entering a novel situation. As already discussed, psychologists agree that low and high SE individuals process data differently and therefore have different cognitive criteria for making self-relevant judgements.  For example, high SE can be defined as the absence of self-doubt and low SE (including narcissism) as the existence of self-doubt,  Upon entering a novel situation, an individual with high SE, and therefore no self-doubt, has only to process data about the situation itself,  The low SE person, on the other hand, is going to have self-doubt evoked (either consciously or subconsciously) and therefore has to process both the situational data AND how they related to their view of themselves, which is a much more complex processing requirement. Narcissistic individuals have an even more complex task in that coping mechanisms like avoidance behavior may not be as acceptable to a narcissist who needs to project a strong image to others. 

An implicit SE test could leverage our understanding of different cognitive strategies to allow researchers to assess the SE status of an individual in a neutral or non-provocative situation.  Take the following scenario:

You enter a bar you’ve never been to before.  It is quite busy and you notice a friend sitting at the end of the bar in conversation with a group of people unknown to you. Your friend does not see you enter.  You would prefer to:

               A) Wait until your friend has seen and acknowledged you before going over and saying hello

               B) Go to the bar and buy a drink before going over to say hello.

              C) Go straight over and say hello.

Option A should appeal more to the low SE group because it gives more time for the complex processing requirements associated with low SE and reduces the activation associated with initiating social contact.  Option B should appeal more to the narcissistic group because it gives a valid reason for taking the time necessary for processing before taking affirmative action, thereby maintaining and reinforcing the “strong” self-image.  Option C would be preferable to the high SE group who do not have the extra cognitive processing requirement or the self-worth triggers in play.

Again, this is a simplistic example designed to demonstrate a point (and note that these specific scenario responses have not been validated).  The point being made is that cognitive criteria can be identified and studied in creative and non-obvious ways to avoid the bias inherent in the current testing modalities.

Perhaps the most complex question relating to the development of a new SE test is the question of validation.  One way of addressing this problem is to develop a large enough data pool.  If sufficient personally relevant in-depth data with specific focus on the day to day experience of living with low or high SE were to be gathered, important differentiating characteristics of SE status would likely emerge. These SE drivers could then be used to develop a test such as the one described above.

Once identified, employing quantitative and objective evaluation of those drivers could be accomplished using electrophysiological or functional imaging endpoints to prospering categorize and validate the concepts.  For example, if high and low SE subjects do indeed process data differently and believed, then it is logical to assume that differences in brain activity would run parallel with these different processing patterns and therefore be measurable and observable.

                                                                                                Conclusion

This discussion has laid out the theoretical basis for the development of an implicit pencil and paper test that could replace general nonspecific inventories like the RSI in the field of SE research.  Considering the fact that the RSI was developed as a superficial test for mass assessment of a generalized psychological phenomenon, it is surprising that no further advancement has arisen in the field.  It is quite likely that such has not happened in part because researchers haven’t fully grasped the fact that, to truly understand human behavior, it is necessary to study the behavior as it is expressed in subjective life.  You cannot gain all of the data required to understand something as complex as SE solely by using blunt tools in the lab.

It is also this author’s opinion that psychologists who study the mental/emotional side of human beings should look more to their neuroscience counterparts.  Mind or brain, it is all one being. Mental and emotional expressions have associated brain activity that can be leveraged to shed more light on complex psychological phenomenon like SE and therefore advance our understanding of the impact SE has on various human behaviors and their relation to mental health.