Assessment of the dimensionality of the Wijma delivery expectancy/experience questionnaire using factor analysis and Rasch analysis

Fear of childbirth has been linked with a number of negative consequences for a woman’s physical and emotional wellbeing. These include pregnancy complications, increased length of labour [1], use of anaesthesia during labour [2, 3], and increased risk of caesarean section deliveries [4, 5]. The content of a woman’s fear may include feelings of lack of control, fear of pain, fear of humiliation, fear for the life and wellbeing of her baby, fear for her own life, and fear of perineal tearing [6]. Many women are fearful of repeating a previous negative birthing experience [7].

One of the most commonly used tools for the measurement of fear of childbirth is the Wijma Delivery Expectancy/Experience Questionnaire (WDEQ-A) which was developed to “measure fear of childbirth by means of the woman’s cognitive appraisal regarding the delivery” [8] (p. 85). Since its development the WDEQ-A has been translated into a number of languages, and has been used in a wide range of studies exploring the correlates and consequences of elevated levels of fear [9‐13]. Although the 33-item WDEQ-A was originally conceptualized as a unidimensional measure, subsequent investigations by other researchers using factor analysis have suggested that it may in fact be multidimensional, tapping a number of different aspects [9, 10, 14, 15]. Johnson and Slade [10] conducted the first factor analysis of the WDEQ-A and concluded that ‘it measured four clear dimensions that are conceptually distinct’ (p.1220). To achieve a satisfactory solution the authors of that study found it was necessary to remove three items (items 26: ‘let happen’, and 28:‘funny’ and item 30: ‘obvious’), with the final four dimensions labeled Fear, Lack of positive anticipation, Isolation and Riskiness.

Australian researchers [14] conducted similar analyses. The four factors defined as ‘fear’, ‘isolation’, ‘lack of positive anticipation’ and ‘riskiness’ in the UK study were similarly identified in the Australian sample, however there were several differences in the actual items included in each factor see Fenwick et al. [14] page 673.

Subsequent analysis undertaken in Norway [16], using both exploratory and confirmatory factor analysis, suggested that a six-factor model showed the best fit, after the removal of eight items. The final solution for the remaining 25-item version of the WDEQ-A included subscales labelled Fear, Negative Appraisal, Loneliness, Lack of self-efficacy, Lack of positive anticipation and Concerns for the child. This six factor structure was later confirmed in a large multi-country European study [12] although no items were removed as none had communalities less than 0.3 [12]. The authors decided that essentially the same factors were found in each of the six European countries and maintained the same factor names as the earlier Norwegian study [16] however these factors were not made up of all the same original items in all the countries [12].

In a study validating an Italian version of the WDEQ-A, [9] the authors raised a number of concerns about the factor analysis procedures undertaken in earlier reports. In particular they noted that in the four factor solutions reported in the literature, items were retained in the scale despite low communalities, failure to load substantially on a single factor, or cross loadings. They also noted that previous researchers had not formally tested the factor structure after removing items from the scale, an important step in determining the factor structure of a scale. Fenaroli and Saita [9] also questioned the validity of retaining a subscale containing only two substantially loading items, labelled Riskiness by Johnson and Slade [10], and Fenwick et al. [14], and Concerns for the child by Lukasse et al. [12] and by Garthus-Niegal et al. [16]. Although these items may conceptually be related to the underlying concept of fear of childbirth, the two items are not sufficient, psychometrically, to form a robust subscale [9]. This concern about the robustness of the two-item Riskiness factor was also raised in a Japanese validation study of the WDEQ-A [15], suggesting that the factor may be “weak and unstable” (p.331). These authors drew attention to the practice of retaining items with relatively low factor loadings and suggested that “more careful attention to the items will be needed in future research” (p.331).

Although many researchers using the WDEQ-A over the past 15 years have calculated a single total score [3, 10, 17] this requires the assumption that the scale is unidimensional. However all authors to date that have tested the dimensionality of the scale have identified distinct factors (between four and six) suggesting multidimensionality [9, 10, 14]. Studies using confirmatory factor analysis have formally tested the appropriateness of a single factor solution and reported very poor fit statistics [9, 16]. Both studies reported comparative fit index (CFI) values below .6, well below the accepted guidelines of .95 for good model fit and .90 for moderate fit [18]. These results, suggesting that the WDEQ-A items do not measure a single underlying dimension, are supported by the low correlations among the factors reported by some authors. Garthus-Niegel et al. for example, reported correlations between the Concerns for the child factor and other WDEQ-A factors ranging from a high of only .298 and a low of .145 [16]. Values this low suggests that this set of items share less than 9% variance with the other factors identified in the WDEQ-A. A summary of studies which have reported results of factor analysis of the WDEQ-A can be found in Table 1.

Low correlations among the factors indicate that women with high scores on one factor do not necessarily have high scores on other factors. For example, just because a woman who feels concern that their child would die or be injured during the labour/birth (item 32, 33) does not necessarily mean that they will feel “lonely” (item 3) or “abandoned” (item 15). The combination of these items to form a single score is clearly inappropriate and may result in the loss of potential information for clinicians in particular. A profile, providing separate subscale scores representing each factor, may prove to be more useful in planning an intervention or providing customized support for an individual woman.

In order to identify and extract a set of subscales from the WDEQ-A that can be used by future researchers and clinicians it is important that the items be subjected to rigorous testing using the latest in psychometric procedures. The importance of good psychometric procedures was emphasized in a recent edition of Journal of Reproductive and Infant Psychology which was devoted to the topic of measurement of psychological health in the prenatal period [19]. These authors suggest that ‘there is much to be gained from new statistical techniques and approaches when developing measures to capture the complexity of psychological health in the perinatal period’ [19] (p. 436). They went on to highlight that: ‘whatever we measure requires rigorous and robust evaluation of the measure both in terms of psychometric standards and interpretation of those standards’ [19] (p.437).

Although the WDEQ-A has been analysed using a number of classical test theory approaches (exploratory factor analysis, confirmatory factor analysis), to date the scale has not yet been subjected to Rasch Analysis, which is based on modern test theory. Rasch analysis is a procedure increasingly being used within the health sciences, medical, psychology and business literatures. It has been used to psychometrically assess many hundreds of clinical tools [20] including the Perinatal Attachment Index [21] and assessment of the birth experience [22]. For a review of the use of Rasch analysis in nursing research, see Hagquist et al. [23]. Unlike classical test theory approaches, Rasch analysis involves a full assessment of all aspects of a scale’s functioning, including its response format, item fit, potential bias, suitability for particular groups, dimensionality and targeting. It allows scales to be refined by removing items that do not fit with the underlying dimension being measured. This assists in the development of short, concise scales that are unidimensional, and are free from item bias.

The use of Rasch analysis to identify and remove WDEQ-A items that do not directly tap the underlying dimension may serve to increase its potential clinical application. In its current form of 33 items the WDEQ-A has been criticized as being too long and complex to be used routinely in clinical settings and to be accurately translated into multiple languages [24]. A recent qualitative study from the United States challenged the utility and appropriateness of the WDEQ-A in its current form for use as a screening tool in a U.S. context where there are many systematic differences in healthcare compared to the Swedish context where the questionnaire was originally developed [25]. In response to this criticism a number of articles have been published recently proposing alternative short measures of childbirth fear [26, 27]. The multidimensional nature of the WDEQ-A, suggested by recent factor analytic studies, may be able to be used to advantage in developing a set of subscales, measuring additional aspects of the construct, over and above the original focus on fear of childbirth.

The aim of this study was to undertake a very detailed psychometric assessment of the WDEQ-A using techniques from both classical test theory (EFA, CFA) and modern test theory (Rasch analysis). The goal was to explore the dimensionality of the scale using EFA and CFA and to identify possible subscales of the WDEQ-A that may have clinical and research utility. Rasch analysis was used to formally assess the response format, suitability of the items, item bias, internal consistency reliability, dimensionality and targeting of the subscales. Additional analyses were also undertaken to explore the correlations among the WDEQ-A subscales and their association with other existing measures, and selected demographic and obstetric characteristics.

This study involved secondary analysis of data from a large Australian randomised control trial designed to test the effectiveness of a midwife led psycho-education intervention to reduce childbirth fear – The BELIEF study [28].

The extensive psychometric evaluation of the WDEQ-A undertaken in this study clearly suggests that it is multidimensional and that it is not appropriate to calculate a total score. In this large sample of 1410 Australian women, four separate subscales were identified (Negative emotions, Lack of positive emotions, Social isolation and Moment of birth), each showing good internal consistency. The pattern of correlations among the WDEQ-A-Revised subscales, and their correlation with other measures (FOBS-The Fear of Birth Scale^TM, EPDS), also suggest that the subscales are measuring different aspects of the underlying concept and therefore should not be combined.

The results of this study are consistent with a growing number of other studies drawing attention to the lack of unidimensionality of the WDEQ-A [9, 10, 14]. The four subscales obtained in the current study are similar, but not identical, to those obtained in other factor analytic studies of the WDEQ-A [9, 10, 14, 15]. The 5-item Negative emotions subscale in the current study was similar in content to a factor labeled Fear by other authors [9, 14]. Our subscale Social Isolation was consistent with items labeled Isolation by Fenwick et al. [14] and Loneliness by Lukasse et al. and Garthus-Niegal et al. [12, 16]. Five of the items in our Lack of positive emotions subscales were also present in the factor labeled by Fenwick et al. [14] as Lack of positive anticipation. Unlike previous researchers we identified a fourth subscale which contains items from the section in the WDEQ-A which asked women “How do you imagine it will feel the very moment you deliver the baby?” This subscale was therefore labeled Moment of birth to distinguish it from the remaining items in the WDEQ-A which asked women to think about how they would feel during the labour and delivery.

This study utilized techniques from both classical test theory (exploratory and confirmatory factor analysis) and modern test theory (Rasch analysis) to fully explore the psychometric properties of the WDEQ-A. This study was the first to utilize Rasch analysis, one of the family of a modern test theory techniques that is now widely being used to explore all aspects of a scale’s performance. It involves assessment of the fit of each of the items, the response scale, item bias, and dimensionality, and allows identification of weak items for removal from the scale. After removal of items based on the application of Rasch analysis, the final 17 -item four-subscale version of the WDEQ-A (referred to as WDEQ-A-Revised) showed good psychometric properties. The removal of poorly performing or unnecessary items resulted in a shorter, more concise measure, better suited to research and clinical utilisation.

Although previous researchers have reported concerns about the multidimensionality of the WDEQ-A, many have continued to calculate a total WDEQ-A score and to use cutpoints based on this score [12, 46]. To derive a total score for a scale it is essential that the items included in the scale are all measuring the same underlying construct [47]. The low intercorrelations among some of the subscales in this study clearly suggest that they are tapping different domains and therefore should not be combined. This is particularly evident for the Social isolation and Moment of Birth subscales which both showed correlations of less than .40 with other subscales. A correlation of .40 represents only 16% overlap in the subscales [32], indicating that respondents with high scores on one subscale, do not necessarily record high scores on other subscales. The combination of subscales with low intercorrelations to create a total score is meaningless.

If the WDEQ-A is not unidimensional, as suggested by this, and a growing number of other studies worldwide, a question is raised concerning its research and clinical utility. How can it be used to assess the level of fear experienced by women facing childbirth? One option is to explore the unique contribution of each of the WDEQ-A-Revised subscales separately. The Negative emotions subscale identified in the current study most closely represents fear of childbirth, containing the items panic, afraid, tense, frightful and weak. Women who consistently endorse above the midpoint on each of these items (representing a mean score of 2.5 or above) may warrant further investigation concerning their perceptions of the upcoming birth. In this study that represents 21.5% of the sample. Elevated scores on the Social Isolation subscale may also be of clinical concern. In our study only 3.8% (n = 54) had a mean score of 2.5 or above on this subscale. Women who feel deserted, alone, lonely or abandoned are potentially at increased risk when facing the challenges of childbirth and motherhood, and may require additional support services, during both the birth and the postpartum period. It is possible that the positive emotions indicated by low scores on the Lack of positive emotions subscale (e.g. confident, relaxed, happy, composed, safe) may serve as a buffer, moderating any potential impact of elevated fear levels. Using the scores from the profile of scores on the Negative emotions, Lack of positive emotions and Social isolation subscales may prove potentially useful in the identification of women requiring counseling, giving top priority to those women with high fear levels that also have a lack of protective positive emotions, and perceived lack of support.

It is unclear how the scores on the Moment of birth subscale could contribute as a either a research or clinical tool, particularly for women who have not yet experienced the birth process. Unlike other items in the scale, which address feelings during both labour and delivery, these items refer specifically to how the woman imagines she will feel the very moment of delivery. This scale recorded the lowest intercorrelation values, both with other WDEQ-A subscales, and other validated tools (EPDS and FOBS^TM). A number of the items in this subscale (eg. Q28 funny, Q30 self-evident) have been identified as difficult for respondents to understand, perhaps due to difficulties in translating from the original Swedish. In a number of previous studies the wording of these items have been modified differently across studies see for example Johnson and Slade and Fenwick et al. [10, 14], making it difficult to compare.

Further investigation of the WDEQ-A-Revised is required to fully understand how each individual subscale may contribute to a better understanding of the emotional health of women facing childbirth. This study was conducted on a large cohort of Australian women, using an English version of the WDEQ-A. Given the widespread use of the scale, further validation of this revised 17 -item, 4-subscale version (WDEQ-A-Revised) is required in different cultural groups. This study focused on the internal validity of the scale and included a very limited range of other measures suitable for exploring the external validation of the tool. Additional studies exploring the association between the WDEQ-A-Revised subscales with other validated tools is needed, and longitudinal studies are required to explore the impact on the birth process and outcome. These investigations will help to determine the unique contribution, and potential usefulness, of the four individual subscales. Additional investigation of the WDEQ-A-Revised is also needed to determine its suitability to clinical practice, particularly given the availability now of shorter, easy to administer, tools such as the FOBS- The Fear of Birth Scale^TM which may be more appropriate in time pressured settings such as hospitals and clinics.

New cutpoints for the WDEQ-A-Revised will need to be established to guide clinicians and researchers in the identification of what constitutes high levels of childbirth fear. The score of >2.5 would suggest that a woman has chosen a response option above the midpoint of the scale on each of the items. To allow comparison of the levels of childbirth fear in future studies the percentage of this sample recording above the midpoint are provided in Table 6. Pending further clinical investigation of this issue it is recommended that women recording above the midpoint (2.5) on each subscale should be considered for further investigation. This could be in the form of further sensitive questioning by the midwife or doctor caring for the woman to understand the content of her fears. By using good clinical judgment, possibly aided by additional screening for anxiety and depression, the woman may be referred to the appropriate resource for more detailed assessment.

The results of this study support the findings of a growing number of other studies that suggest that the WDEQ-A is multidimensional and therefore a total scale score should not be calculated. The four subscales identified in the current study may provide researchers with a more refined, psychometrically sound tool to explore the concept of fear of birth, and the differential impact of the various aspects included in the WDEQ-A-Revised. Further testing is needed to assess the revised format of the WDEQ-A in different cultural settings, and to evaluate its potential applicability in clinical settings to evaluate the emotional health of women facing childbirth and identify those at risk.