The use of composite multi-item scales to measure latent (i.e., unobservable) constructs is widespread in survey research across the disciplines. Yet, the length of these scales (many upwards of 15 items) poses challenges for survey administration: high survey costs, increased respondent burden, and item non-response (Coste et al. 1997; Stanton et al. 2002; Smith, Combs, and Pearson 2012). To address these challenges, researchers seek to define and use abbreviated scales (see, for example, Blumberg et al. 1999; Levine 2013).
Although shortening multi-item scales is common practice, as Goetz et al. (2013) point out, the strategies scholars use in the shortening process often lack “methodological rigor,” calling the validity of these abbreviated measures into question (p. 711). To address this, over the years, researchers have issued a series of methodological guidelines suggesting best practices for scale shortening (see, for example, Coste et al. 1997; Smith, McCarthy, and Anderson 2000; and Stanton et al. 2002 as cited in Goetz et al. 2013).
The emphasis of such guidelines is typically focused on the first phase of the shortening process: defining an abbreviated scale. Much less attention is given to the second phase of the shortening process: validation. To the extent that it is given attention, researchers widely recommend that the abbreviated scale be validated on an independent sample (Coste et al. 1997; Smith, McCarthy, and Anderson 2000; Stanton et al. 2002; Smith, Combs, and Pearson 2012; Goetz et al. 2013; Kruyen, Emons, and Sijtsma 2013; Sitarenios 2022). Yet, the guidance stops short of recommending the use of split-ballot experiments, the gold-standard technique in evaluating question wording differences (Schuman and Presser 1981). Further, little guidance outlines how to validate a categorical abbreviated scale (but see Smith, McCarthy, and Anderson 2000) and no guidelines, to our knowledge, provide empirical illustrations of successful validation exercises.
Therefore, in this paper, we provide a step-by-step empirical illustration of scale shortening that includes both phases of the shortening process. We begin by illustrating how to define a shortened version of a scale by following several agreed-upon recommended practices as described in the literature. We then illustrate how a split-ballot experiment can be used to validate an abbreviated categorical composite score. In doing so, this paper also provides an illustration of how to thoroughly document and justify decisions made throughout the process, a move widely recommended across the guidelines (see, for example, Smith, McCarthy, and Anderson 2000; Goetz et al. 2013).
Our motivating example is the abbreviation of the Transportation Security Index (TSI) from a 16-item measure to a 6-item measure. The TSI is a validated measure of transportation insecurity, a condition in which an individual is unable to regularly move from place to place in a safe or timely manner due to an absence of resources necessary for transportation (Gould-Werth, Griffin, and Murphy 2018; Murphy, Gould-Werth, and Griffin 2021). Modeled after the Food Security Index (National Research Council 2006), the TSI was designed to measure transportation insecurity at the individual level based on the way people experience it qualitatively, regardless of mode of transit or geography. The TSI has been cited as a useful measure of transportation-related material hardship (Murphy et al. 2022), a valuable evaluation tool (Sung et al. 2023), and a potentially useful screening tool for clinicians (Brandt et al. 2023). Yet, as Turner et al. (2020) have pointed out, its 16-item length is burdensome and cost prohibitive for inclusion on most questionnaires, warranting the development of an abbreviated form.
To identify and validate an abbreviated TSI, we drew upon data derived from surveys and cognitive interviews.
Survey data. Survey data were gathered from two similar data collections administered in May 2018 and November 2022. The 2018 survey was fielded to validate the original TSI-16 (see Murphy, Gould-Werth, and Griffin 2021) and to develop a preliminary abbreviated scale. Accordingly, all respondents (analytic sample size = 1,999) were administered the full TSI-16. The 2022 survey was fielded to validate the proposed abbreviated scale and included a split-ballot experiment wherein one random half-sample (analytic sample size = 1,099) received the original TSI-16 and the other random half-sample (analytic sample size = 1,118) received the proposed abbreviated scale (See Appendix A for the 2022 survey questionnaire; items comprising the abbreviated scale are in bold font). Each survey was administered to a distinct sample of Ipsos’ (formerly GfK Group) KnowledgePanel® members. Recognizing that the unique transportation behaviors of college-aged young adults might impact our results, we restricted both survey samples to U.S. adults aged 25 years or older. Both surveys also included oversamples of respondents living in households at or below the federal poverty line. For further details about each of our data collection efforts, including information about the KnowledgePanel® and descriptive statistics of each sample, please refer to Appendix B.
Cognitive interview data. In 2015, to identify the initial pool of candidate TSI items, we conducted 52 cognitive interviews with a socioeconomically and demographically diverse group of respondents in Chicago and urban, suburban, and rural Michigan (see Gould-Werth, Griffin, and Murphy 2018). These cognitive interviews were again considered here. Respondents were identified through nonprofit organizations, door knocking, and snowball sampling. During the interview, respondents were administered our candidate items, probed to assess comprehension, recall, and judgement, and asked about their financial and transportation situations.
Methods & Results
In this section, we provide a step-by-step illustration of how to define and validate an abbreviated version of a categorical composite scale, using the TSI as an example. For each step, we begin by providing a methodological justification for the step, noting the recommended guidelines where they exist. We then detail how, for each step, we implemented these practices in the shortening of the TSI. Throughout this discussion, all survey data were weighted and analyzed using either Stata 15.1 (StataCorp 2017) or Mplus 6.1 (Muthén and Muthén 1998–2010).
Defining a shortened version of a scale
Step 1: Document the validity and measurement properties of the original scale. The methodological guidelines for shortening scales broadly agree that only those original scales that have been validated and demonstrated to have good measurement properties should be shortened (see, for example, Coste et al. 1997; Smith, McCarthy, and Anderson 2000; Goetz et al. 2013). Because the abbreviated scale should preserve (or improve upon) the original scale’s psychometric properties, it is important to first document the psychometric properties (e.g., dimensionality, validity, reliability) of the original scale from which the abbreviated scale will be derived. As Goetz et al. (2013) argue, doing so enables potential users of the abbreviated scale to better understand how decisions around shortening were made.
In our case, the original scale is the validated 16-item Transportation Security Index (TSI-16). The TSI-16 measures an individual’s experience with transportation insecurity by asking how often (never = 0, sometimes = 1, often = 2) in the past 30 days respondents have experienced 16 unique symptoms of transportation insecurity observed in qualitative research (see Table 1). Symptoms fall into two categories that prior psychometric analyses (Murphy, Gould-Werth, and Griffin 2021) demonstrate are indicators of a single latent trait (i.e., transportation insecurity; Cronbach’s = 0.95): (1) material symptoms that reflect the difficulties people have getting from place to place in a safe or timely manner (e.g., skipping trips, arriving places late) and (2) relational symptoms that reflect the emotional toll and social strain of experiencing transportation insecurity (e.g., being embarrassed, worrying about inconveniencing ride givers).
The development of the TSI-16 was the result of a multi-step process. As described in Gould-Werth, Griffin, and Murphy (2018), item content was informed by extensive qualitative research, including 187 interviews. A preliminary index was identified using exploratory factor analysis on survey data collected in 2016 (Gould-Werth, Griffin, and Murphy 2018). This index was then validated on a different nationally representative survey sample (administered in 2018) by using confirmatory factor analysis and other analytic methods (Murphy, Gould-Werth, and Griffin 2021). Used as a categorical measure, the TSI-16 identifies five categories of transportation insecurity generated from an individual’s sum score (0-2 = secure, 3-5 = marginal, 6-10 = low, 11-16 = moderate, 17-32 = high insecurity) (McDonald-Lopez et al. 2023).
Step 2: Define an objective for the abbreviated scale. Methodological guidelines widely recommend that the objectives for defining an abbreviated scale be made explicit at the outset of the shortening process, and that they include the anticipated benefits to be derived from an abbreviated scale as well as how many items will be needed for this shortened scale to meet these goals (see, for example, Smith, McCarthy, and Anderson 2000; Goetz et al. 2013). Documenting such information is important not only because the defined objectives shape item selection and other methodological considerations, but also because, as Goetz et al. (2013) write, providing such information will help potential users of an index decide whether the original or shortened version of a scale should be administered.
With this in mind, we defined four objectives for our abbreviated TSI. First, taking our conceptual model into account as recommended by Goetz et al. (2013), we wanted the abbreviated scale to efficiently capture both the material and relational manifestations of transportation insecurity (content validity) most likely to be encountered across a variety of survey contexts, including those with relatively smaller sample sizes. Second, we wanted the abbreviated scale to have face validity among both respondents and researchers. Face validity for respondents would increase respondent motivation and thus the quality of data collected. Face validity for researchers would facilitate the use of the scale in research. Third, we desired a categorical abbreviated scale that would demonstrate concordance with the type of transportation insecurity categories defined by the categorical original scale. Finally, given that empirical work using the TSI has focused on quantifying the prevalence of transportation insecurity (Murphy et al. 2022), we aimed to develop an abbreviated TSI that would capture transportation security’s prevalence as precisely as the original scale does. Recognizing the generally low prevalence of the most severe categories of transportation insecurity (e.g., 3% and 5% of U.S. adults were estimated to experience high and moderate transportation insecurity, respectively [Murphy et al. 2022]) and the likelihood of the measure being dichotomized in external analyses, we privileged items that distinguished between respondents experiencing transportation security and respondents experiencing any level of insecurity.
We did not identify a specific target length that would be needed to meet these objectives. We did, however, desire to identify a scale that had no fewer than three items, the minimum number of items required for a one-factor model.
Step 3: Use both content and statistical approaches to select items and document the item selection process. Detail the justification for item retention or removal, including whatever tradeoffs were made in such decisions. The literature suggests that it is a best practice to ensure that the abbreviated scale retains the psychometric properties of the original by using statistical approaches to evaluate what items should be retained or struck (see, for example, Coste et al. 1997; Smith, McCarthy, and Anderson 2000; Stanton et al. 2002; Goetz et al. 2013; Sitarenios 2022). Because it is also important to preserve the content validity of the original scale, methodological guidelines also widely recommend simultaneously taking the content of each item into account when conducting such an evaluation (see, for example, Coste et al. 1997; Stanton et al. 2002; Smith, McCarthy, and Anderson 2000; Goetz et al. 2013).
Following this logic, we approached shortening the TSI by considering what individual items we could justifiably discard. Evaluating the psychometric properties of each item (“statistical approach”), we began by ranking all 16 items by their item discrimination and item difficulty parameters (“never to sometimes”) as estimated by a graded response model using our 2018 survey data (see Table 2). Graded response models estimate the probability that a respondent will endorse a particular item response given the respondent’s location on a latent continuum (here, transportation insecurity), the ability of the item to differentiate among respondents at different locations on the latent continuum (item discrimination), and the location on the latent continuum at which the respondent has a 50 percent chance of endorsing a particular item response (item location). A desirable set of items will have high discrimination values while adequately covering the content space (i.e., including easier and more difficult items) (DeVellis 2017; Sitarenios 2022).
Recognizing that individuals experiencing the greatest level of transportation are less likely to be detected in applications with smaller sample sizes, we first removed the most difficult item to endorse (avoiding). Next, although paying what Lowe and Mosby (2016) call the “time tax” is central to the experience of transportation insecurity, our results showed that the four items related to time (late, took longer, early, waiting) were the least discriminating, likely because transportation secure people also perceive themselves to incur travel time costs (McDonald-Lopez et al. 2023). Although the recommended guidelines for shortening scales emphasize the importance of preserving the content validity of the original scale, such considerations must be weighed against the fact that any abbreviated scale must only retain items that most efficiently differentiate those experiencing transportation insecurity from those who are transportation secure. Because these items do not accomplish this objective and because we are retaining other items that tap into the material dimension of insecurity, we elected to remove them.
Given that our statistical approach did not suggest striking any additional items, we drew on our cognitive interview data to evaluate the performance of each of our remaining 11 items (“content approach”). Analysis revealed that when thinking about feeling bad, respondents considered feelings related to feeling left out and embarrassed. Because feeling bad encompassed the two items that respondents interpreted more narrowly, thus producing semantic redundancy, we struck left out and embarrassed (see Stanton et al. 2002 for a discussion of eliminating items based on semantic redundancy). Similarly, we removed not invited, keeping the more general and all-encompassing relationship effects.
We decided to retain not able to leave house when you want to over stuck – items capturing a similar experience – for two reasons. First, admitting to “feeling stuck at home” might be perceived as stigmatizing by some respondents, potentially resulting in their disengagement from the response task. Such an item would thus undermine our objective of identifying an abbreviated scale that would increase respondent motivation. Second, in addition to capturing people who are stuck at home, not able to leave the house when you want to also captures the lack of autonomy that transportation insecure people experience when they have to rely on the schedules and reliability of public transit and social networks for rides and thus covers more symptoms associated with transportation insecurity.
Although “worry” questions have worked well in indices measuring other forms of material hardship, like food insecurity, our evaluation of respondent comprehension indicated that, in some cases, respondents interpreted worry overly broadly, to include, for example, concerns about inconveniences related to traffic or road construction. For this reason, we struck worry.
Ultimately, then, six items – 3 material and 3 relational – were retained for the abbreviated TSI, preserving the content validity of the original scale: reschedule, skipped, not able to leave house when you want to, felt bad, inconvenience, and relationship effects (see Table 2; TSI-6 items are in bold font).
Validating the abbreviated version of a scale
The validation of the abbreviated scale helps determine the extent to which the abbreviated scale preserves (or improves upon) the psychometric properties of the original scale, a necessary requirement of an effective abbreviated scale (Coste et al. 1997; Smith, McCarthy, and Anderson 2000; Goetz et al. 2013; Kruyen, Emons, and Sijtsma 2013; Sitarenios 2022). Below, we describe the way we structured each step of the validation process, from data collection to analysis, in an effort to ensure a rigorous comparison of our abbreviated and original scales, thus demonstrating how to provide a convincing validation of an abbreviated scale.
Step 1: Conduct a split-ballot experiment on an independent sample using the same data collection procedures and sample design used in validating the original scale. To decrease the likelihood that the abbreviated scale would be overfitted to a particular sample, the literature recommends testing abbreviated indices on new, independent samples representing the same target population (see, for example, Coste et al. 1997; Smith, McCarthy, and Anderson 2000; Stanton et al. 2002; Smith, Combs, and Pearson 2012; Goetz et al. 2013; Kruyen, Emons, and Sijtsma 2013; Sitarenios 2022). More specifically, we recommend a split-ballot survey design wherein the original scale is administered to one random half-sample and the abbreviated scale to the other random half-sample. Such a technique is used widely to compare the effectiveness of question wording alternatives (see Schuman and Presser 1981) and is well suited for comparing different versions of measurement scales because it ensures that comparisons between the original and abbreviated scales are not conflated with any difference in the survey sample or data collection procedures. A split-ballot design also protects against “halo effects” which occur when only the original scale is administered and abbreviated items are extracted from it or when both the original and abbreviated forms are administered to the same sample in the same survey, two common practices in the literature (Goetz et al. 2013). In such designs, responses to the abbreviated scale are likely influenced by the concurrent administration of the remaining original scale items, thus impacting the generalizability of the results.
Accordingly, in 2022, we fielded a new survey on an independent sample. We administered the original TSI-16 to one random half-sample (“Ballot One”) and the abbreviated TSI-6 to the other random half-sample (“Ballot Two”). To minimize the variability in comparisons across survey efforts due to differences in survey methods, in 2022, we contracted with the same firm (Ipsos) and used the same panel (Knowledge Panel®) as we used in our 2018 survey. We also used the same sampling parameters (i.e., adults over age 25 and an oversample of those below the poverty line).
Step 2: Evaluate the consistency of the original scale over time. In order for the proposed abbreviated scale to accurately represent the original scale, it is important to first determine that the original scale performs as expected in the new independent sample.
Because the reproduction of prevalence estimates is one of the objectives for our abbreviated scale, in our case, we compared prevalence estimates derived from the TSI-16 in 2018 and 2022 (Ballot One only, by definition). As illustrated in Table 3, prevalence estimates across the five categories of transportation insecurity did not meaningfully vary. Thus, 2018 and 2022 data are comparable and an abbreviated scale derived from the 2022 data that performs as well as the original scale measured in 2022 should, on its face, also represent the original scale validated in 2018.
Step 3: Evaluate the psychometric properties of the abbreviated scale. To evaluate whether the abbreviated scale preserves the original scale’s psychometric properties, consider examining the abbreviated scale’s dimensionality, reliability, and concurrent validity (which, for a categorical scale is assessed in steps 5 and 6) (Coste et al. 1997; Smith, McCarthy, and Anderson 2000; Stanton et al. 2002; Goetz et al. 2013; Sitarenios 2022).
Previous research demonstrated that the material and relational manifestations of transportation insecurity, as measured by the TSI-16, are best reflected by a single construct (i.e., transportation insecurity) (Murphy, Gould-Werth, and Griffin 2021). To evaluate whether the dimensionality of the abbreviated scale replicates that of the original scale, we used confirmatory factor analysis to conduct a nested model comparison using Ballot Two data. Specifically, we compared the more restrictive one-factor model in which the correlation between the material and relational factors is constrained to be equal to one to a two-factor model in which the correlation between the two factors is freely estimated. Although the restricted model resulted in a significantly worse model fit (χ2(1)=7.600, p<.001), the estimated correlation between the two factors was 0.983, which equates to 96.6% shared variance. Therefore, following the principle of parsimony (see also DeVellis 2017), a one-factor model, with a high level of internal consistency (Cronbach’s = 0.92), is supported, demonstrating that the TSI-6 preserves two key psychometric properties of the original scale.
Step 4: Create cut points for the abbreviated scale using data from respondents in the new sample who were administered the abbreviated scale. To evaluate whether the abbreviated scale reproduces prevalence estimates derived from the categorical original scale, abbreviated scale categories, or cut points, first need to be identified. This can be achieved using similar methods as were used in creating cut points for the original scale.
In our case, we conducted a k-means cluster analysis using data from Ballot Two (abbreviated scale only) respondents. In this non-deterministic partitional clustering method, observations are iteratively clustered into k mutually exclusive and exhaustive categories using their continuous TSI sum scores as input (MacQueen 1967). Generally, smaller values of k will result in solutions that are more reproducible; however, meaningful substantive differences between observations might be missed. Therefore, we desired to identify a k which provided as much description of the population as could be generally reproduced. Given our prior identification of a five-category TSI-16 (secure, marginal, low, moderate, high insecurity), we determined that between three and five distinct categories of transportation insecurity might be identified using the abbreviated scale. Accordingly, we estimated k=3, k=4, and k=5 means clustering models. Because the method is nondeterministic (i.e., results could differ each time the model is estimated), we re-estimated each model 10 times.
As illustrated in Table 4, among the 3-, 4-, and 5-cluster solutions estimated, only the 3-cluster solution exhibited consistent replication across a majority of iterations. This solution identified three clusters defined by sum scores of 0–1 (secure), 2–5 (marginal/low), and 6–12 (moderate/high). These clusters thus define our preliminary three-category abbreviated scale.
Step 5: Calculate the level of agreement between the original and abbreviated scales. Such calculations (e.g., percent agreement, Kappa statistic) determine the extent to which the categorical abbreviated scale aligns with the categorical original scale (i.e., concurrent validity) (Coste et al. 1997; Smith, McCarthy, and Anderson 2000; Stanton et al. 2002; Goetz et al. 2013).
In our case, because the number of categories between scales differed, we began by examining the distribution of the five TSI-16 original scale categories across the continuous TSI-6 abbreviated scale sum scores among only Ballot One respondents. As expected, the percentage of respondents classified as “secure” (value 1) (per the original scale) decreases as the abbreviated scale sum score increases (see Figure 1). Furthermore, the pattern suggests that the three categories identified in the abbreviated scale closely resemble a collapsed original scale categorization: The first categories of both the abbreviated and original scale generally identify respondents who are transportation secure. The second category of the abbreviated scale (sum scores between 2 and 5, inclusive) primarily identifies respondents who experience marginal or low insecurity (per the original scale; values 2 and 3). Finally, the third category of the abbreviated scale (sum scores between 6 and 12, inclusive) primarily identifies respondents who experience moderate or high insecurity (per the original scale; values 4 and 5).
To more formally estimate the concordance between the categorical original and abbreviated scales, we calculated the percent agreement between the two using the 2022 survey data. As illustrated in Table 5, 90.8 percent (weighted) of all respondents completing Ballot One were similarly classified across both forms: 78.2 percent as transportation secure between scales, 6.7 percent as experiencing marginal or low insecurity, and 5.9 percent as experiencing moderate or high insecurity.
Because the simple percent agreement between two measures does not take into account chance agreement, we next estimated the Kappa statistic between the three-category abbreviated scale and the three-category original scale that was created by collapsing the original five categories as discussed above (i.e., 1=1, 2=2,3, 3=4,5). As estimated on the Ballot One sample, the Kappa statistic was 0.76, reflecting substantial (Landis and Koch 1977) or excellent (Fleiss, Levin, and Paik 1981) agreement.
Step 6: Use chi-square analysis to compare prevalence estimates derived from the original and abbreviated scales. Because the performance of an abbreviated categorical scale depends on its ability to classify people in the same way the original scale does (Smith, McCarthy, and Anderson 2000; Smith, Combs, and Pearson 2012; Kemper et al. 2018; Sitarenios 2022), prevalence estimates derived from the original and abbreviated scales should be compared. To do this, create a single x-category variable across the entire new data set such that respondents who received the ballot with the original scale are assigned their x-category original scale score and respondents who received the abbreviated scale are assigned their x-category abbreviated scale score. Next, to determine whether there is a significant difference in prevalence estimates between the two scales, conduct a chi-square analysis.
Toward that end, we created a single three-category TSI variable across the entire 2022 data set such that Ballot One respondents were assigned to one of three categories defined by the original scale score cut points, and Ballot Two respondents were assigned to one of three categories defined by the abbreviated scale score cut points. We then conducted a weighted chi-square analysis which revealed no significant difference in prevalence estimates between the two scales (see Table 6; design-based F(1.99, 4406.47)=1.7910, p=0.167). There is initial evidence, then, that the TSI-6 is a sufficient proxy for the TSI-16 when estimating transportation insecurity’s prevalence.
This paper presented the steps we took to define and validate the TSI-6. By doing so, we aimed to provide readers with a useful empirical illustration of how to define and validate an abbreviated categorical scale in line with some of the best practices in survey research. It is our hope that such an illustration also provides a useful example of how to thoroughly document and justify all decisions and considerations made throughout the shortening process. As the methodological guidelines recommending such transparency note, providing such documentation is important because it provides potential users of the abbreviated measure with the information they need to evaluate its strengths and weaknesses and whether they wish to use it (Smith, McCarthy, and Anderson 2000; Goetz et al. 2013).
Our example was the shortening of the 16-item Transportation Security Index (TSI-16). Using nationally representative survey data and cognitive interview data and drawing upon statistical and content approaches, we developed and validated the TSI-6: six questions that can be used to determine one’s level of transportation insecurity. Importantly, this abbreviated scale met our objectives as outlined at the beginning of the paper: (1) the scale captures both the material and relational manifestations of transportation insecurity, (2) the items have face validity, (3) the scale identifies comparable categories of insecurity as the original scale, and (4) the scale generates comparable prevalence estimates as the original scale. Therefore, the TSI-6 can be used to achieve parsimony with little loss of information. Moreover, it can do so while decreasing respondent burden and survey costs. Based on our 2022 survey data, whereas the median time to complete the TSI-16 was 2.12 minutes, the median time to complete the TSI-6 was just under 1 minute.
Importantly, the specific processes we followed were iterative and dependent on the unique properties of our scale, our research objectives, and the results each step garnered. For example, our example involved shortening a unidimensional scale. There are many composite scales, however, that have multiple factors, each of which must be preserved in the shortening process (Smith, McCarthy, and Anderson 2000; Goetz et al. 2013). Our example also involved validating a single defined abbreviated scale. There are other cases, however, where researchers might be considering multiple abbreviated scale options. In these cases, we would recommend an experiment including one ballot for the original scale and one ballot for each of the proposed abbreviated scales. Finally, because we aimed to develop an abbreviated scale that would capture transportation insecurity’s prevalence as precisely as the original scale does, our validation efforts placed special emphasis on comparing how the abbreviated scale performed against the original scale with respect to prevalence. Other researchers may have additional objectives for their abbreviated scales which should guide their validation efforts (Sitarenios 2022). For instance, those who are interested in preserving the predictive validity of their original scale will want to add an additional step to their validation efforts: a comparison of how the defined abbreviate scale compares with the original scale in predicting some outcome of interest (Stanton et al. 2002).
Depending on the broader research objectives of a study, using an abbreviated scale might not always be preferred to using the original scale. As is the case with many survey design decisions, the tradeoffs must be carefully weighed. For example, it might not be worth decreasing the overall survey length or reducing other survey costs, when the psychometric properties of the abbreviated scale are worse than those of the original scale (Kemper et al. 2018). Furthermore, as in our example presented here, the abbreviated scale might identify a coarser categorization of the latent construct than does the original scale. To the extent that greater differentiation of respondents is desired, questionnaire space is available, and survey sample sizes are sufficient, the original scale may be the preferred measure.
Of course, measurement development is an ongoing process. As with the development and validation of original scales, once an abbreviated scale has been validated, researchers should seek to replicate their findings in different survey contexts, examining how the abbreviated scale performs with different modes of administration, target populations, and questionnaire contexts.
Alexandra K. Murphy
Department of Sociology
University of Michigan
3115 LSA Building
500 South State Street
Ann Arbor, MI 48109
We thank Mike Bader and David Pedulla for providing feedback during the early stages of our work defining the abbreviated TSI. We are also grateful to the following agencies whose financial support made this publication possible: National Science Foundation (grant OIA09936884); the Stanford Center on Poverty and Inequality (grant H79AE000101 from the US Department of Health and Human Services); and the University of Michigan’s Poverty Solutions and Mcity initiatives, College of Literature, Science, and the Arts, Office of Research, and Department of Sociology. Any opinions, findings, and conclusions or recommendations expressed in this article are those of the author(s) and do not necessarily reflect the views or official policies of the National Science Foundation or the US Department of Health and Human Services.