Use of Item Response Models in Assessing the Health Literacy Facet Understanding Health Information for Early Childhood Allergy Prevention and Prevention of COVID-19 Infections by Pregnant Women and Mothers of Infants

Health Literacy – Definition and Models

According to Nutbeam’s (2000) definition, health literate people have cognitive, social, and motivational skills and abilities that enable them to identify, process, and understand health information and services to make appropriate decisions and act on that information. Functional HL refers to literacy skills and abilities required for validly reading and understanding health information (including numerical skills; Numeracy; Nutbeam, 2000; Sørensen et al., 2012). Communicative or Interactive HL is required for goal-oriented behavior in the healthcare system, and Critical HL enables people to appropriately evaluate health information (Nutbeam, 2000). In addition to literal skills, health-related knowledge is defined as a further fundamental component of HL in Soellner and colleagues’ model (2010). Particular importance is attached to motivational and action-oriented facets, which play a major role in determining the health-literate actions of individuals. Sørensen et al. (2012) assume that the successive steps of Access, Understand, Appraise, and Apply must be regarded as central facets of HL. All HL models share in common that (1) HL must be understood as a multidimensional construct and (2) literate abilities and skills – in the sense of Functional HL – represent fundamental model facets as well as determinants of health-literate behavior. The models differ primarily regarding the consideration of further psychological and systemic constructs as well as assumptions about decision-making and behavioral processes (Mackert et al., 2015; Soellner & Rudinger, 2018; Sørensen et al., 2012).

Developing an Assessment for the HL Facet Understand Health Information

In this article, methods of competence structure and level modeling are used for the investigation and measurement of the HL facet Understand Health Information (Klieme et al., 2008). Although this is only one facet of the HL construct, it is defined as a distinct subconstruct and is included in all models as a fundamental requirement for health-literate information processing. The facet Understand is most likely to be considered in the sense of Functional HL (Nutbeam, 2000). Furthermore, a valid understanding of information is a necessary prerequisite for health-literate information appraisal and decision-making (Paasche-Orlow & Wolf, 2007). Since an achievement test is to be developed for the living situation of new parents, a clear distinction has to be made here – independent of personal and social moderator effects – between correct and incorrect results of cognitive information processing. Because of the clarity of the response scoring and the definition of distinct elementary processing requirements using single items, this opens up a solid basis for transferring the standards of competence modeling (item response models) to the field of health competence modeling (Klieme et al., 2008; Pant, 2013; Schmuckle & Egloff, 2011).

Competencies or literacy traits are generally defined as personal behavioral dispositions that can be actualized or utilized purposefully in information processing as well as in planning and controlling behavior. Competencies determine the successful solution or accomplishment of performance requirements or tasks (Weinert, 2001). By definition, competencies are application-oriented and domain-specific (Weinert, 2001). Competence levels are determined in particular by acquired knowledge and learning experiences. Accordingly, a valid survey on HL has to be situation- and context-related to ensure that parents’ daily life conditions and the ensuing demands are considered. To meet these requirements and to be able to investigate domain specificity of Understand Health Information, the present study strived to develop parental HL items based on information materials targeted to parents for two content areas: early childhood allergy prevention (ECAP; Dresch et al., 2021) and COVID-19 infection prevention (COVID-19-IP; Cluver et al., 2020).

Health information materials can be considered a medium intended to induce certain cognitive changes in the recipients (e. g., enhanced problem awareness, knowledge or problem-solving skills) and, consequently, to improve their health-related decision-making and actions. Health-literate people possess the cognitive abilities to extract the right information from health information or draw the right conclusions from the materials for themselves. Faulty understanding occurs when information is derived from materials that contradict the intentions of the media designer. However, erroneous understanding is not necessarily caused by erroneous processing; the media content itself can be systematically error-prone or induce misconceptions (Andrulis & Brach, 2007).

Accordingly, the validity of the content and the didactic design must first be ensured (Wilson, 2005). The following aspects must be considered for an objective performance measure of Understand Health Information (Dresch et al., 2021):

1. 1.
   Does the information material belong in the defined content area (ECAP, COVID-19-IP)? Is it representative of the content area?
2. 2.
   Does the material aim to change cognitive representations in people who do not yet have the appropriate knowledge (especially misconceptions)?
3. 3.
   Is the information to be conveyed clearly defined?
4. 4.
   Is the information to be conveyed evident or suitable to validly increasing health-related knowledge?
5. 5.
   Is the presentation format appropriate to conveying the information adequately? Can erroneous processing result from the design of the representation?
6. 6.
   Is the information material suitable for the target group? What prior knowledge is assumed? Are the information needs and the life situation of the target group taken into account?
7. 7.
   Can response options be identified that clearly indicate whether the content to be conveyed was validly processed / understood by the respondents (target responses) or was incorrectly processed / understood (distractors)?

Psychometric Modeling and Measurement Standards for Competence / Literacy Assessment

Particularly in the field of educational research, methods of competence structure and competence level modeling based on the assumptions of item response theory (IRT; Rupp et al., 2010; van der Linden, 2021) represent the diagnostic gold standard for analyzing competency domains and capturing competence levels (Wilson, 2005). However, methods for assessing HL do not yet use these advantageous psychometric standards (Haun et al., 2014). Subjective assessments (e. g., European Health Literacy Survey Questionnaire [HLS-EU-Q], Sørensen et al. 2013) capture the self-assessed ability to identify and process health information and to decide and act on this basis. Measures obtained by self-assessments systematically deviate empirically and conceptually from objective indicators of competence because of individual information processing (e. g., individual reference systems, self-serving bias; Schmuckle & Egloff, 2011). Existing objective achievement tests (e. g., Test of Functional Health Literacy in Adults [TOFHLA]; Parker et al., 1995) are generally limited to basal literacy skills (word recognition, reading comprehension, numeracy). Additionally, the psychometric foundations of existing objective test procedures are limited to basal validations according to the classical test theory and limited psychometrically reasonable evaluation rules (Eid & Schmidt, 2014).

The present paper examines whether the basic HL facet Understand Health Information can be modeled as a latent trait dimension by items based on informational materials targeting pregnant women and mothers of infants. IRT models (van der Linden, 2001; Wilson, 2005) assume that the solution probability p‍(+) results in a logistic transformation of the difference in ability θ_v of a woman ν, the difficulty σ_i of the task i to be solved, and its discrimination α_i:

If the discrimination strength is identical for all items (α_I = 1; VAR‍(α_i) = 0), this model is called the 1-pl or the Rasch model. In the case of the 2-pl or Birnbaum model, the discriminatory strengths of the items are allowed to vary and are accordingly estimated item-specifically (VAR‍(α_i) ≠ 0). Item groups satisfying the IRT assumptions form an ideal basis for analyzing the dimensional structure of literacy domains, since the assumption that the underlying latent dimensions – in this case, Understand Health Information – correspond unambiguously with the empirical data information is tested empirically. If it is assumed that not only one but multiple competence dimensions or facets determine the solution of tasks, multidimensional IRT models (mIRT; Rupp et al., 2010) can be used.

The central aim of this study is to test the psychometric and dimensional structure of an item pool for Understand Health Information. Using IRT methods, we investigate the following questions based on test items developed for the application areas of ECAP and COVID-19-IP by pregnant women and mothers of infants.

1. 1.
   Is there a common latent HL dimension Understand Health Information underlying the processing of the test items?
2. 2.
   Does the model fit improve by distinguishing the prevention contents (1) ECAP, (2) COVID-19-IP, and (3) General Health Prevention (GHP)?
3. 3.
   Which items represent the underlying HL dimension‍(s) according to the assumptions of IRT?
4. 4.
   Are the identified HL levels of Understand Health Information associated with sociodemographic and allergy- and COVID-19-related characteristics of pregnant women and mothers of infants?

Item Construction

The definition of item contents was based on a comprehensive literature and internet search on health information in ECAP and COVID-19-IP. The focus was on information directly targeted to parents in the area of ECAP and adults in general in COVID-19-IP (e. g., www.allergiecheck.de, www.quarks.de, www.gesundheitsinformation.de, www.rki.de).

Additionally, 13 parents and 5 students conducted an internet-based search on health-related information on ECAP (average duration approximately 120 min; range: 41 – 252 min). The screen contents and the continuous accompanying verbalization of parental thoughts (think-aloud method; Pohontsch & Meyer, 2015) were recorded. In a subsequent guided interview (average duration: 43 min, range: 35 – 60 min), the parents reflected in particular on their prior knowledge and information needs, their search strategies, and their assessment of the quality and usefulness of the identified information materials. Based on this, an item pool with 67 test items was defined and item-specific response options for the multiple-choice answer format (correct answers and distractors corresponding to typical comprehension errors) were used.

Each item consists of an item stem (graph, table, text) conveying information and answer alternatives, one or more of which apply in terms of correct understanding (for example items, see Electronic Supplements 1 and 2). For 22 items, multiple choice answers were coded dichotomously (“0” = wrong, “1” = correct) because each incorrect selection of any of the 4 – 6 individual choice options indicated a fundamental misunderstanding of the information presented. One (.5) or two (.33 and .67) partial credits were defined for 29 and 16 items, respectively, because the answers can indicate correct understanding despite the partially incorrect response elements. Of the total 67 test items of the assessment instrument, 29 refer to the content area ECAP and 30 to COVID-19-IP. Eight test items refer to GHP topics for (expectant) parents. In 18 subsequent pretesting sessions (think aloud technique), we optimized the item bank regarding content validity, suitability, and comprehensibility (incl. optimization of distractor responses).

Data Collection and Study Sample

Data collection was conducted using an online questionnaire to obtain a sufficiently large and appropriate sample despite the limitations imposed by the COVID-19 pandemic. The data collection was divided into three parts (about 45 minutes each; interval: about 10 days). Part 1: Sociodemographic, allergy-‍, and health-related characteristics (e. g., self-efficacy, risk competence); Part 2: ECAP-HL items; Part 3: COVID-19-IP-HL items. Each participant was randomly assigned a test booklet containing 12 – 14 items. Details of the survey using test booklets are given in Electronic Supplement 3.

Data collection took place online from May 2021 to February 2022. The survey design and acquisition of participants were guided by the Tailored Design Method (TDM; Dillman et al., 2015). Participants were free to complete the HL items independently or as part of an individual digital video session. The Electronic Supplement 3 contains detailed information on the recruitment strategy. All 11 project staff members involved in recruitment had been trained prior to the start of recruitment and contacted participants in a standardized manner (e. g., providing identical information about study organization and participation).

Initially, we aimed to generate the sample entirely with parent couples. However, it turned out that the willingness of male partners to participate was limited. Hence, in the second half of the sampling period we decided to continue to recruit only women. Furthermore, because women with lower educational status were underrepresented, especially women without university entrance qualifications were enrolled subsequently. Thus, we succeeded in recruiting a sufficiently large sample size of women that appropriately represents the German population regarding educational status. With a proportion of 53 % of women with higher education entrance qualifications, the sample accurately reflects the school education situation in the overall female population in Germany in the age range 25 – 44 years (Bundesinstitut für Bevölkerungsforschung, 2021). For the final data analysis, we included only women: A mixed sample of couples and individuals would have implied that stochastic dependence within couples could only have been partially accounted for by hierarchical linear modeling and would have posed a particular problem because of the small number of participating male partners.

In the end, we included a total of 343 mothers (pregnant women: n = 62; 18 %; mothers with children aged 0 – 3 years: n = 281; 82 %) in the study sample; n = 170 (52 %) are primipara. Their socioeconomic status as measured by the McArthur Scale (Hoebel et al., 2015) is slightly higher in the sample (M = 5.87; SD = 1.42) than in the reference standard sample of women aged 18 – 44 years (M = 5.45; SD = 1.50) (Hoebel et al., 2015). The average age of the respondents was 33.3 years (range: 20 – 47; SD = 4.6). Most of the women (n = 193, 57 %) were affected by at least one allergy, whereas most of the children (n = 291, 84 %) did not suffer from any allergy to date. The sample characteristics are shown in detail in Electronic Supplement 4.

Statistical Analysis

We conducted item and structural analyses using exploratory (EFA) and confirmatory factor analysis (CFA) for categorical, ordinal data, and IRT methods (restricted maximum likelihood [RML]; Harville, 1977) implemented in MPLUS 8.3. RML allows valid estimation in case of missing completely at random (MCAR), which was ensured because of the random assignment of test booklets by design. The comparison of models with different complexity (1-pl vs. 2-pl model) or dimensionality is based on measures from information theoretic measures (Rost & Langeheine, 1997), which allow an integrated assessment of model complexity and data fit (Akaike information criterion [AIC]; Bayesian information criterion [BIC]). We performed the final IRT analysis using Conquest software (Adams et al., 2020), using the multidimensional random coefficient multinominal logit (MRCML) approach, a multidimensional extension of the 1-pl or Rasch model. To test the fit of the items to the model assumption (logistic dependence of the solution probabilities on the latent trait), we calculated weighted and unweighted deviation measures (mean square deviations, MNSQ). The model-consistent items had a value of about 1. Values higher than 1 (underfit) would indicate that responses to this item contain more stochastic information than the underlying model would predict; item fit values below 1 would indicate that the item fits better than predicted by the model (overfit). The latter usually occurs because of local dependencies of the items, which lead to higher item associations than expected based on latent trait expression. Thus, items with fit values < 1 do not pose a problem for the item validity as an indicator of latent trait and can be accepted as scale items (van der Linden, 2021). To avoid systematically penalizing larger sample sizes in IRT-based tests, items are usually not selected according to the criterion of significance. Instead, items with fit measures < 1.20 are considered acceptable (or 1.25 in the Pisa study; Frey et al., 2006). The quality of the overall test (group of all model-compliant items) is further assessed using the separation reliability (values > .7 acceptable; > .8 good; van der Linden, 2021). The person-specific estimates of the person parameters are obtained using the weighted-likelihood-estimate (WLE).

Analysis of the Dimensional Structure Underlying the HL Items

EFA identified the single-factor structural assumption (1–DIM) as the best fitting for the pool of 67 items according to BIC (difference to the two-factor solution: ΔBIC = -271.70; see Electronic Supplement 5; Burnham & Anderson, 2002). Confirmatory IRT analysis also revealed the most appropriate data fit for the 1-DIM model, which assumes only one latent dimension, Understand (BIC [1-pl] = 16,542.97). The AIC value, whose statistical validity is inferior to the BIC, showed a slightly better value for the 2-DIM-2pl model (ΔAIC = -7.59). However, in the 2-DIM-2pl model, the very high latent correlation r = .93 of the two dimensions ECAP / GHP and COVID-19-IP also indicates the homogeneity of the item pool.

Thus, the unidimensional structure proves to be superior, although the items on COVID-19-IP not only refer to a different content area but were also measured at a different points in time (data collection at intervals of approximately 10 days on average). Accordingly, the construct intercorrelation of r = .93 in the 2-DIM-2pl model may also be interpreted in terms of construct stability or retest reliability over the test interval. In this respect, it is noticeable that the item group in the GHP domain should theoretically be similarly associated with the items on ECAP and COVID-19-IP. But the latent correlation of GHP to the simultaneously measured content area ECAP is r = .999 and thus considerably higher than the latent correlation of GHP and COVID-19-IP: r = .93. Both the 1- and 2-factor model structures are further examined as plausible latent structural model alternatives.

Competing model definitions that distinguish the need to process number vs. number-independent information (numeracy; Kumar et al., 2010) or the presentation formats text, graphics, and tables (Nitsch et al., 2014) do not provide an adequate model alternative. In addition to the poorer values of the information-theoretic measures (see Electronic Supplement 5), the high latent construct correlations (r = .993 / .997 and r = .980 – 1.000, respectively) indicate the psychometric identity of the assumed subfacets.

For all model definitions, BIC indicates the superiority of the more parsimonious 1-pl- / Rasch model over the 2–pl- / Birnbaum model (see Electronic Supplement 5). Thus the validity of analyzing the item pool according to the assumptions of the 1-pl model is warranted.

Analysis of Item and Scale Properties

The results of the analysis of the item pool using the MRCML estimation implemented in CONQUEST are shown in Table 1. In the first model estimations, the items A08, A09, A12, A30, C04, and C24 showed significant positive deviations from the expected value of 1 in the (un)‌weighted fit. Although all other items exhibited values below the critical value of 1.20 in the first estimation, the fit values of C09, C13, C21, and C19 rose beyond the critical value > 1.20 as a result of the stepwise item elimination. Finally, after excluding these 10 items, weighted fit values for the remaining 57 items did not deviate significantly upwards from 1 and did not exceed the critical value of 1.20. Unweighted fit values for a few single items were slightly above this limit (up to max. 1.25). For the final pool of 57 items, the one- and two-dimensional (ECAP / ‌GHP vs. COVID-19-IP) models yield similar model fits (ΔAIC = 9.70 and ΔBIC = 2.02 each less than 10; Anderson & Burnham, 2002; see Electronic Supplement 5).

Table 1 Item characteristics based on confirmatory 1pl-IRT-analysis for the 67 items assessing the HL facet Understanding Health Information (MCMRL estimation; CONQUEST)

ECAP/GHP							COVID-19-IP
Item	hi(resp.)a)	σib)	λic)	ritd)	MNSQ		Item	hi (resp.)a)	σib)	λic)	ritd)	MNSQe)
	0/1/2/3				Unw. | Weig.			0/1/2/3				Unw. | Weig.
A11	.13/.17/.70	-1.11	.54	.49	1.06 | 1.10		C05	.43/.33/.24	0.36	.43	.44	1.25 | 1.17
A06	.06/.25/.69	-1.49	.50	.45	1.12 | 1.03		C09g)	.11/.55/.33				(1.26* | 1.26*)
A41	.04/.20/.75	-1.80	.38	.41	1.21 | 1.00		C13g)	.39/.28/.33				(1.30* | 1.20*)
A17	.05/.56/.39	-1.16	.64	.52	0.90 | 0.91		C22	.37/.35/.29	0.12	.58	.56	0.96 | 0.98
A38	.23/.77	-1.28	.67	.51	0.82 | 0.87		C26	.25/.23/.38/.14	0.20	.58	.60	1.05 | 1.06
A14	.30/.36/.34	-0.01	.52	.56	1.03 | 1.05		C30	.48/.30/.23	0.48	.49	.48	1.12 | 1.04
A22	.28/.44/.28	0.07	.32	.37	1.23 | 1.21		C01	.30/.38/.32	-0.04	.51	.54	1.07 | 1.04
A33	.15/.85	-1.96	.64	.42	0.78 | 0.94		C14	.14/.86	-2.10	.09	.09	1.55 | 1.15
A37	.27/.73	-1.09	.38	.32	1.07 | 1.05		C20	.17/.23/.43/.16	-0.04	.60	.62	1.03 | 1.02
A30f)	.14/.14/.32/.40				(1.29* | 1.23*)		C24f)	.27/.11/.28/.34				(1.56* | 1.35*)
A36	.17/.22/.61	-0.94	.64	.63	0.81 | 0.95		C28	.16/.43/.41	-0.63	.39	.54	1.00 | 1.02
A27	.36/.64	-0.62	.60	.51	0.90 | 0.92		C10	.33/.31/.25/.11	0.55	.53	.60	1.05 | 1.04
A42	.31/.69	-1.06	.58	.49	0.89 | 0.92		C03	.36/.19/.35/.10	0.68	.58	.65	1.10 | 1.00
A02	.29/.71	-1.20	.75	.63	0.71 | 0.81		C15	.44/.27/.29	0.39	.65	.65	0.85 | 0.89
A09f)	.12/.53/.28				(1.25* | 1.29*)		C18	.13/.32/.55	-0.96	.65	.56	0.95 | 0.97
A18	.08/.27/.65	-1.58	.50	.47	1.08 | 1.04		C23	.33/.32/.36	0.01	.55	.56	1.08 | 1.04
A39	.14/.44/.42	-0.84	.59	.57	0.99 | 0.99		C27	.16/.53/.32	-0.42	.55	.53	0.98 | 0.99
A29	.26/.74/	-1.34	.52	.47	0.92 | 0.96		C02	.31/.36/.26	0.79	.59	.61	1.04 | 1.04
A07	.20/.27/.54	-0.83	.55	.56	1.06 | 1.08		C08	.07/.16/.27/.56	-0.90	.58	.57	1.04 | 1.04
A12f)	.30/.52/.18				(1.28* | 1.27*)		C11	.68/.25/.07	1.63	.67	.50	0.84 | 0.86
A24	.26/.74	-1.35	.71	.60	0.76 | 0.83		C17	.16/.16/.67	-1.07	.68	.62	0.85 | 1.00
A35	.26/.74	-1.35	.60	.51	1.02 | 0.90		C21g)	.33/.29/.25/.13				(1.44* | 1.30*)
A34	.19/.81	-1.83	.65	.51	0.83 | 0.90		C25	.23/.16/.61	-0.74	.72	.70	0.73 | 0.85
A04	.23/.38/.39	-0.35	.53	.57	1.01 | 1.01		C04f)	.47/.35/.17/.02				(1.35* | 1.25*)
A08f)	.25/.52/.24				(1.29* | 1.28*)		C07	.22/.78	-1.54	.14	.30	1.21 | 1.02
A15	.18/.30/.52	-0.79	.60	.60	0.94 | 0.97		C12	.28/.72	-1.19	.43	.37	0.96 | 0.98
A31	.23/.71	-1.10	.56	.49	0.91 | 0.96		C16	.19/.32/.50	-0.77	.44	.51	1.23 | 0.99
A19	.40/.60	-0.52	.54	.48	1.01 | 0.98		C19g)	.30/.29/.32/.09				(1.33* | 1.24*)
A21	.29/.27/.44	-0.33	.63	.62	0.89 | 0.94		C29	.44/.40/.16	0.61	.50	.50	1.05 | 1.02
G01	.07/.93	-2.80	.14	.10	1.24* | 1.07		C06	.28/.23/.25/.25	0.13	.57	.64	1.07 | 1.08
G10	.17/.34/.49	-0.63	.59	.53	1.00 | 1.01
G06	.12/.88	-2.26	.75	.51	0.63 | 0.89
G09	.09/.91	-2.53	.79	.38	0.59 | 0.91
G03	.14/.86	-2.28	.51	.38	0.94 | 0.98
G07	.15/.85	-2.17	.42	.32	0.99 | 1.03
G05	.14/.28/.58	-1.12	.57	.57	1.13 | 0.99
G08	.65/.35	0.67	.34	.31	1.15 | 1.06

Note. GHP = General Health Prevention, ECAP = Early Childhood Allergy Prevention, COVID-19-IP = COVID-19 infection prevention ^a) h_i‍(resp.) = relative frequencies of response categories: “0” = wrong, “1” = correct for dichotomous items; else “partial credit”; “2” = correct for three-level items; else “partial credit”; “3” = correct for four-level items; ^b) σ_i = item parameter (logit); ^c) λ_i = factor loading; ^d) r_it = corrected item-total correlation; ^e) MNSQ = mean square deviation (weighted / unweighted); ^f) weighted MNSQ > 1.20 in first estimation step (all items); ^g) weighted MNSQ > 1.21 (t-value = 2.0) in subsequent estimation steps after deletion of other misfitting items. * = Significant deviation of MNSQ from the expected value 1 ( p < .05).

Table 1 Item characteristics based on confirmatory 1pl-IRT-analysis for the 67 items assessing the HL facet Understanding Health Information (MCMRL estimation; CONQUEST)

View as image HTML

The 57 items conforming to the model assumptions have good seperation reliability of .86, with a person variance on the latent trait of 0.95. The Wright map in Figure 1 shows the distribution of the person and item parameters graphically. The estimated z-standardized person parameters (women’s ability) are depicted at the upper part of Figure 1. The distribution of the item parameters (σ) = -0.76; range: -2.80 – 1.63, see bottom of Figure 1) is shifted to the negative relative to the distribution of the person parameters (ϑ) = -0.02 range: -2.64 – 3.16). Thus, the item pool is rather “easy” regarding the ability of the participating women. Accordingly, the solution probabilities for the items tend to be above 50 %.

Association of Person Parameters on the Trait Understand Health Information with Further Women’s Characteristics

Person-parameter estimates of women’s capability to Understand Health Information are positively related to school graduation (r = .46) and self-assessed socioeconomic status (r = .27; Table 2). The person parameter is weakly correlated with women’s allergy status (r = .11), occurred COVID-19 infection (r = .12), informedness about COVID-19 (r = .12), perceived burden by the pandemic (r = –.11), and the statement that everything should be done to fight the pandemic (r = .13).

Table 2 Correlations of IRT person-parameters in Understand Health Information with parents’ sociodemographic variables as well as allergy- and COVID-19-specific person characteristics (N = 343)

Sociodemographic variables
Age	.172***
Number of children	-.060
Age of the youngest child in months	.039
School graduationa)	.456***
Subjective socioeconomic stateb)	.271***
Allergy-specific person characteristics
Parents affected by allergyc)	.114*
Child affected by allergyc)	.054
Self-assessed impairment because of allergiesd)	.127
COVID-19-specific person characteristics
Already infectedc)	.120*
Infection with severe progression in the personal environmentc)	.089
I have regularly and extensively informed myself about the subject of COVID-19e)	.120*
I felt burdened by COVID-19e)	-.107*
I had a fear of a COVID-19 infectione)	-.030
It is important that everything possible be done to fight the COVID-19 pandemice)	.129*

Note.^a) Response categories: “0” = no university entrance qualification, “1” = University entrance qualification or higher; ^b) McArthur-Scale: 1 = low to 10 = high; ^c) response categories: “1” = no, “2” = yes; ^d) “Please indicate to what extent you feel impaired by your allergy‍(s) or allergic disease‍(s) in your everyday life in general?”, response categories: “1” = very little / not at all to “6” = very strong; ^e) high values = high agreement. ^*p < .05, ^**p < .01, ^***p < .001.

Table 2 Correlations of IRT person-parameters in Understand Health Information with parents’ sociodemographic variables as well as allergy- and COVID-19-specific person characteristics (N = 343)

View as image HTML