Measurement instrument development

The instrument (Figure 4.1) was designed to display and to be assessed on a number of psychometric properties, which in combination indicate a robust measure (Anastasi & Urbina, 1997). The properties sought and the design steps taken to attain and test them are described below.

Content validity is the comprehensive assessment of all domains of interest for the instrument; in this case, the spectrum of attitudinal and demographic factors known to be relevant to parents’

MMR decision-making. This property was attained by including a question on each of the 20 attitudes and seven demographics on which MMR-accepting parents and MMR-rejecting parents have been shown to significantly and consistently differ. These attitudes and demographics were identified through recent systematic reviews of relevant quantitative and qualitative evidence (Brown, Kroll et al, 2010a; Roberts, Dixon-Woods et al, 2002; Falagas & Zarkadoulia, 2008; Mills, Jadad et al, 2005), and through a recent large UK epidemiological study of demographic

predictors of MMR intention/uptake (Pearce, Law et al, 2008). The items span attitudes toward five key dimensions: MMR vaccine, measles infection, social influences, information sources, and practical barriers. Attainment of this property was qualitatively tested through feedback from expert members of the project team (not reported).

120 Figure 4.1: Measurement instrument

121 Internal consistency is the capacity of the instrument to obtain internally correlated responses to individual items which are intended to measure a shared broader construct. It is psychometrically desirable to group items together in scales as this reduces the possibility of measurement

artefacts or outlying responses on a single item skewing the data. Grouping items also increases statistical power and provides clearer direction for intervention design. This property was attained by grouping together items which are conceptually linkedand which have been

correlated elsewhere in the literature (Brown, Kroll et al, 2010a). Attainment of this property was statistically tested by obtaining Cronbach’s alpha coefficients for each of the scales.

Test-retest reliability is the capacity of the instrument to obtain consistent responses at two timepoints between which the attitudes assessed would not be expected to differ. This was attained through a) item selection and phrasing designed to tap stable ‘evidence-based’

cognitions rather than more changeable emotional states, and b) simple low-literacy

questionnaire design including an intuitive and familiar five-point response scale, coupled with some reverse-scored items to encourage careful consideration of each one rather than rote answering. Attainment of this property was statistically assessed by administering the instrument to a subset of respondents at two timepoints and correlating responses.

Concurrent validity is the capacity of the instrument to distinguish between groups which are expected to differ on the measured items/scales. This was attained by assessing attitudes in the instrument on which parents have been shown to differ by their child’s MMR status. Attainment of this property was statistically tested by comparing responses to each individual attitude item and each attitude scale between three participant subgroups: those whose child received MMR on-time, those whose child received MMR late, and those whose child did not receive MMR.

Predictive validity is the capacity of the instrument to predict an outcome which is expected to be predictable from the domains measured. This was attained again by assessing attitudes in the instrument which have been previously shown to relate to MMR intention/uptake. Attainment of

122 this property was statistically tested by entering all items assessed in the measurement

instrument (both individually and as scales) into hierarchical logistic regressions with child’s MMR status as the outcome.

4.2.2.3 Procedure

The Health Protection Agency classified the study as a service evaluation not requiring ethical approval. Consent to participate was implied through study completion. A copy of the instrument was posted to the parent/guardian of every child in the sample, along with a cover letter (Figure 11.8, Chapter 11 Appendices) explaining the purpose and provenance of the study, a freepost return envelope, and a translation sheet advising (in seven languages most commonly used in the PCTs: Arabic, Farsi, Polish, Punjabi, Somali, Tamil and Urdu) that translations were available on request. At 3 and 6 weeks after the first copy was sent, non-respondents received a reminder letter with another copy of the instrument and enclosures, to a maximum of two postal reminders per household. At 4 and 7 weeks after the first copy was sent, a telephone reminder was administered to non-respondents with a working telephone number provided by the PCT: a total of four attempts per reminder (2 daytime, 2 evening) were made, to a maximum of two answered calls or answerphone messages per household. During telephone reminders an interpreter was available, and participants were invited to respond to the instrument during the call rather than return their copy by post. A subset of 225 participants who responded to the first copy of the instrument with no reminders were sent a test-retest duplicate with enclosures and a cover letter explaining the purpose of the duplicate, approximately 3-6 weeks after they returned their first copy; no reminders were sent in relation to the test-retest duplicates.

123 CHIS-recorded MMR dose history at 5^th birthday, and postcode-level Indices of Multiple

Deprivation 2007 data (IMD2007 (Department of Communities and Local Government, 2007)), were obtained for the entire sample when every child had turned 5 years old. Where MMR dose history obtained at the end of the study period differed from that provided at the start, the most up-to-date history was used. For analysis, MMR dose history was categorised in line with HPA COVER methodology (Health Protection Agency, 2010) and the UK Department of Health immunisation schedule (Department of Health, 2007) during the study period: for MMR dose 1 analyses, “on-time” was defined as dose received up to and including child’s 2^nd birthday, “late”

was defined as dose received after child’s 2^nd birthday up to and including child’s 5^th birthday, and

“none” was defined as no dose received by day after child’s 5^th birthday. For MMR dose 2 analyses, “on-time” was defined as dose received after 3 years 4 months (recommended age for MMR dose 2 in immunisation schedule) up to and including child’s 5^th birthday, and “none” was defined as no dose received by day after child’s 5^th birthday. Free-text responses to the job item were coded by two independent analysts (good agreement between analysts: Cohen’s

Kappa=0.91) to the 8-class version of the National Statistics Socio-Economic Classification (NS-SEC (Office for National Statistics, 2005)), where code 1 is the highest socio-economic class (higher managerial/higher professional/large employer) and code 8 the lowest (never worked/long-term unemployed/student etc); respondents classifying themselves as ‘mother’, ‘housewife’ or similar were coded to category 8.

4.2.2.4 Analysis

Data were analysed using SPSS v 17.0 (SPSS Inc). Response rates and respondent characteristics were assessed using all available data on the entire sample. Test-retest reliability analyses (Pearson’s r correlation coefficients) and tests of internal consistency of predefined attitude scales (Cronbach’s alpha coefficients) were conducted using the raw data. Subsequently, missing

124 values were imputed using within-participant scale means for scales of 5 items or more where up to 2 items were missing. Scale scores were calculated by summing scores (including imputed values) for individual items comprising the scale then dividing by the number of items in the scale.

In univariate analyses with nominal outcomes (parent ethnicity, parent marital status, MMR status category), Chi-square tests (unadjusted) and ordinal regression (adjusted for child age and IMD2007 score) were used. Ordinal outcomes (child age, parent age, education, number of children, occupation) were compared using Mann-Whitney tests and Kruskal-Wallis tests (unadjusted) and ANCOVA (adjusted as before). Scale outcomes (all attitude items, child age, IMD2007 score) were compared using independent samples t-tests and ANOVA (unadjusted) and ANCOVA (adjusted as before). Post-hoc pairwise comparisons (time vs late, late vs none, on-time vs none) were conducted for all three-group analyses using Mann-Whitney tests, Z tests, and post-hoc main effects comparison with Bonferroni correction for multiple comparisons.

Hierarchical logistic regressions were used to identify and quantify independent predictors of receipt and non-receipt of each MMR dose (late recipients of MMR dose 1 were excluded from these analyses). Predictors were entered in three steps starting with those expected, based on the literature, to be most predictive: first were objective demographics, then subjective (self-report) demographics, and finally attitudes.

4.2.3 Results