Should a Normal Imputation Model be Modified to Impute Skewed Variables?

Hippel, Paul T. von

doi:10.1177/0049124112464866

Cited by 72 publications

(87 citation statements)

References 42 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In addition, one of the problems when imputing the scores at the scale level was the asymmetric distribution of the scores and, although transforming is recommended 45 to get better imputed values, transformation can yield substantial bias if the transformed variable is not close to normal. 46 We therefore considered two alternative approaches to how the outcome of family satisfaction was imputed:…”

Section: Discussionmentioning

confidence: 99%

Family-Reported Experiences Evaluation (FREE) study: a mixed-methods study to evaluate families’ satisfaction with adult critical care services in the NHS

Wright

Walmsley

Harvey

et al. 2015

Health Serv Deliv Res

View full text Add to dashboard Cite

BackgroundTo improve care it is necessary to feed back experiences of those receiving care. Of patients admitted to intensive care units (ICUs), approximately one-quarter die, and few survivors recollect their experiences, so family members have a vital role. The most widely validated tool to seek their views is the Family Satisfaction in the Intensive Care Unit questionnaire (FS-ICU).ObjectivesTo test face and content validity and comprehensibility of the FS-ICU (phase 1). To establish internal consistency, construct validity and reliability of the FS-ICU; to describe family satisfaction and explore how it varies by family member, patient, unit/hospital and other contextual factors and by country; and to model approaches to sampling for future use in quality improvement (phase 2).DesignMixed methods: qualitative study (phase 1) and cohort study (phase 2).SettingNHS ICUs (n = 2, phase 1;n = 20, phase 2).ParticipantsHealth-care professionals, ex-patients, family members of ICU patients (n = 41, phase 1). Family members of ICU patients (n = 12,303, phase 2).InterventionsNone.Main outcome measuresKey themes regarding each item of the 24-item FS-ICU (FS-ICU-24) (phase 1). Overall family satisfaction and domain scores of the FS-ICU-24 (phase 2).ResultsIn phase 1, face validity, content validity and comprehensibility were good. Adaptation to the UK required only minor edits. In phase 2, one to four family members were recruited for 60.6% of 10,530 patients (staying in ICU for 24 hours or more). Of 12,303 family members, 7173 (58.3%) completed the questionnaire. Psychometric assessment of the questionnaire established high internal consistency and criterion validity. Exploratory factor analysis indicated new domains:satisfaction with care,satisfaction with informationandsatisfaction with the decision-making process. All scores were high with skewed distributions towards more positive scores. For family members of ICU survivors, factors associated with increased/decreased satisfaction were age, ethnicity, relationship to patient, and visit frequency, and patient factors were acute severity of illness and invasive ventilation. For family members of ICU non-survivors, average satisfaction was higher but no family member factors were associated with increased/decreased satisfaction; patient factors were age, acute severity of illness and duration of stay. Neither ICU/hospital factors nor seasonality were associated. Funnel plots confirmed significant variation in family satisfaction across ICUs. Adjusting for family member and patient characteristics reduced variation, resulting in fewer ICUs identified as potential outliers. Simulations suggested that family satisfaction surveys using short recruitment windows can produce relatively unbiased estimates of average family satisfaction.ConclusionsThe Family-Reported Experiences Evaluation study has provided a UK-adapted, psychometrically valid questionnaire for overall family satisfaction and three domains. The large sample size allowed for robust multilevel multivariable modelling of factors associated with family satisfaction to inform important adjustment of any future evaluation.LimitationsResponses to three free-text questions indicate the questionnaire may not be sensitive to all aspects of family satisfaction.Future workReservations remain about the current questionnaire. While formal analysis of the free-text questions did not form part of this proposal, brief analysis suggested considerable scope for improvement of the FS-ICU-24.Study registrationCurrent Controlled Trials ISRCTN47363549.Funding detailsThe National Institute for Health Research Health Services and Delivery Research programme.

show abstract

Section: Discussionmentioning

confidence: 99%

Family-Reported Experiences Evaluation (FREE) study: a mixed-methods study to evaluate families’ satisfaction with adult critical care services in the NHS

Wright

Walmsley

Harvey

et al. 2015

Health Serv Deliv Res

View full text Add to dashboard Cite

show abstract

“…Similarly, von Hippel [14] found truncated normal regression resulted in biased inference. For data with a weak skew, the truncated regression method performed well, particularly in the MCAR scenario.…”

Section: Discussionmentioning

confidence: 99%

“…As post-imputation rounding is conducted after the imputed datasets have been created, there are concerns that the rounding of imputed values may cause bias in the resulting parameter estimates, particularly for the marginal mean of the imputed variable, and will inappropriately reduce the variance of the imputed values [14]. …”

Section: Introductionmentioning

confidence: 99%

Comparison of methods for imputing limited-range variables: a simulation study

Rodwell

Lee

Romaniuk

et al. 2014

BMC Med Res Methodol

View full text Add to dashboard Cite

BackgroundMultiple imputation (MI) was developed as a method to enable valid inferences to be obtained in the presence of missing data rather than to re-create the missing values. Within the applied setting, it remains unclear how important it is that imputed values should be plausible for individual observations. One variable type for which MI may lead to implausible values is a limited-range variable, where imputed values may fall outside the observable range. The aim of this work was to compare methods for imputing limited-range variables, with a focus on those that restrict the range of the imputed values.MethodsUsing data from a study of adolescent health, we consider three variables based on responses to the General Health Questionnaire (GHQ), a tool for detecting minor psychiatric illness. These variables, based on different scoring methods for the GHQ, resulted in three continuous distributions with mild, moderate and severe positive skewness. In an otherwise complete dataset, we set 33% of the GHQ observations to missing completely at random or missing at random; repeating this process to create 1000 datasets with incomplete data for each scenario.For each dataset, we imputed values on the raw scale and following a zero-skewness log transformation using: univariate regression with no rounding; post-imputation rounding; truncated normal regression; and predictive mean matching. We estimated the marginal mean of the GHQ and the association between the GHQ and a fully observed binary outcome, comparing the results with complete data statistics.ResultsImputation with no rounding performed well when applied to data on the raw scale. Post-imputation rounding and imputation using truncated normal regression produced higher marginal means than the complete data estimate when data had a moderate or severe skew, and this was associated with under-coverage of the complete data estimate. Predictive mean matching also produced under-coverage of the complete data estimate. For the estimate of association, all methods produced similar estimates to the complete data.ConclusionsFor data with a limited range, multiple imputation using techniques that restrict the range of imputed values can result in biased estimates for the marginal mean when data are highly skewed.

show abstract

“…Whichever method is chosen to account for missing data, it is important to understand why and how the values got missing in the first place ( [44]; [40]; [43]). Missing data can be thought of as being caused in one or some combinations of ways which [20] outlined as; random processes, processes that are measured, and process that are not measurable.…”

Section: Assumptions About Missing Datamentioning

confidence: 99%

Challenges and Implications of Missing Data on the Validity of Inferences and Options for Choosing the Right Strategy in Handling Them

Dibal¹

2017

IJSDA

View full text Add to dashboard Cite

Abstract:Missing data in surveys and experimental research is a common occurrence which has serious implications on the validity of inferences. Advances in statistical procedures provides better and efficient methods of handling missing data yet many researches still handle incomplete data in ways that affects the results negatively. We review in detail the mechanisms that generates missingness, and the appropriate methods to account for the missing values to enable the researcher have adequate knowledge to make informed decision on the choice of method to account for missingness.

show abstract

Should a Normal Imputation Model be Modified to Impute Skewed Variables?

Cited by 72 publications

References 42 publications

Family-Reported Experiences Evaluation (FREE) study: a mixed-methods study to evaluate families’ satisfaction with adult critical care services in the NHS

Family-Reported Experiences Evaluation (FREE) study: a mixed-methods study to evaluate families’ satisfaction with adult critical care services in the NHS

Comparison of methods for imputing limited-range variables: a simulation study

Challenges and Implications of Missing Data on the Validity of Inferences and Options for Choosing the Right Strategy in Handling Them

Contact Info

Product

Resources

About