Amount Of Missing Data Research Articles

Sample Size Requirements for Structural Equation Models: An Evaluation of Power, Bias, and Solution Propriety.

Determining sample size requirements for structural equation modeling (SEM) is a challenge often faced by investigators, peer reviewers, and grant writers. Recent years have seen a large increase in SEMs in the behavioral science literature, but consideration of sample size requirements for applied SEMs often relies on outdated rules-of-thumb. This study used Monte Carlo data simulation techniques to evaluate sample size requirements for common applied SEMs. Across a series of simulations, we systematically varied key model properties, including number of indicators and factors, magnitude of factor loadings and path coefficients, and amount of missing data. We investigated how changes in these parameters affected sample size requirements with respect to statistical power, bias in the parameter estimates, and overall solution propriety. Results revealed a range of sample size requirements (i.e., from 30 to 460 cases), meaningful patterns of association between parameters and sample size, and highlight the limitations of commonly cited rules-of-thumb. The broad "lessons learned" for determining SEM sample size requirements are discussed.

Effects of missing data on species tree estimation under the coalescent

With recent advances in genomic sequencing, the importance of taking the effects of the processes that can cause discord between the speciation history and the individual gene histories into account has become evident. For multilocus datasets, it is difficult toachieve complete coverage of all sampled loci across all sample specimens, a problem that also arises when combining incompletely overlapping datasets. Here we examine how missing data affects the accuracy of species tree reconstruction. In our study, 10- and 100-locus sequence datasets were simulated under the coalescent model from shallow and deep speciation histories, and species trees were estimated using the maximum likelihood and Bayesian frameworks (with STEM and *BEAST, respectively). The accuracy of the estimated species trees was evaluated using the symmetric difference and the SPR distance. We examine the effects of sampling more than one individual per species, as well as the effects of different patterns of missing data (i.e., different amounts of missing data, which is represented among random taxa as opposed to being concentrated in specific taxa, as is often the case for empirical studies). Our general conclusion is that the species tree estimates are remarkably resilient to the effects of missing data. We find that for datasets with more limited numbers of loci, sampling more than one individual per species has the strongest effect on improving species tree accuracy when there is missing data, especially at higher degrees of missing data. For larger multilocus datasets (e.g., 25–100loci), the amount of missing data has a negligible effect on species tree reconstruction, even at 50% missing data and a single sampled individual per species.

Human motion capture data recovery using trajectory‐based matrix completion

Matrix completion has previously been used as a method to recover missing data in a mocap sequence. Improvments to this scheme by structuring motion capture data into an alternative representation for matrix completion are proposed. This representation takes advantage of the properties of mocap data and matrix completion to improve the reliability of matrix completion. Experiments show that the proposed motion capture data recovery method performs significantly better for nearly all tested motion sequences and amounts of missing data.

Recommendations for managing missing data, attrition and response shift in palliative and end-of-life care research: Part of the MORECare research method guidance on statistical issues

Background: Statistical analysis in palliative and end-of-life care research can be problematic due to high levels of missing data, attrition and response shift as disease progresses. Aim: To develop recommendations about managing missing data, attrition and response shift in palliative and end-of-life care research data. Design: We used the MORECare Transparent Expert Consultation approach to conduct a consultation workshop with experts in statistical methods in palliative and end-of-life care research. Following presentations and discussion, nominal group techniques were used to produce recommendations about attrition, missing data and response shift. These were rated online by experts and analysed using descriptive statistics for consensus and importance. Results: In total, 20 participants attended the workshop and 19 recommendations were subsequently ranked. There was broad agreement across recommendations. The top five recommendations were as follows: A taxonomy should be devised to define types of attrition. Types and amount of missing data should be reported with details of imputation methods. The pattern of missing data should be investigated to inform the imputation approach. A statistical analysis plan should be pre-specified in the protocol. High rates of attrition should be assumed when planning studies and specifying analyses. The leading recommendation for response shift was for more research. Conclusions: When designing studies in palliative and end-of-life care, it is recommended that high rates of attrition should not be seen as indicative of poor design and that a clear statistical analysis plan is in place to account for missing data and attrition.

Analyzing weight loss intervention studies with missing data: Which methods should be used?

ObjectiveMissing data due to study dropout is common in weight loss trials and several statistical methods exist to account for it. The aim of this study was to identify methods in the literature and to compare the effects of methods of analysis using simulated data sets. MethodsLiterature was obtained for a 1-y period to identify analytical methods used in reporting weight loss trials. A comparison of methods with large or small between-group weight loss, and missing data that was, or was not, missing randomly was conducted in simulated data sets based on previous research. ResultsTwenty-seven studies, some with multiple analyses, were retrieved. Complete case analysis (n = 17), last observation carried forward (n = 6), baseline carried forward (n = 4), maximum likelihood (n = 6), and multiple imputation (n = 2) were the common methods of accounting for missing data. When comparing methods on simulated data, all demonstrated a significant effect when the between-group weight loss was large (P < 0.001, interaction term) regardless of whether the data was missing completely at random. When the weight loss interaction was small, the method used for analysis gave considerably different results with mixed models (P = 0.180) and multiple imputations (P = 0.125) closest to the full data model (P = 0.033). ConclusionThe simulation analysis showed that when data were not missing at random, treatment effects were small, and the amount of missing data was substantial, the analysis method had an effect on the significance of the outcome. Careful attention must be paid when analyzing or appraising studies with missing data and small effects to ensure appropriate conclusions are drawn.

SNA data difficulties with dark networks

Social network analysis (SNA) conclusions are drawn on terrorist and dark network data sets that may provide erroneous results due to an indeterminate amount of missing data or data corruption. Compounding these effects, information sources reporting on terrorist groups and other dark network organizations may intentionally or unintentionally provide false data. These introduced errors may be significant as they could produce analytic results that are counter to the true situation, leading to misappropriation of resources, improper strategy adoption, and erroneous actions. Analyst cognizance of the causes of imperfect social network data, the importance of proper boundary specification, biases introduced via the employed data collection methods, and characteristics of social network information sources, particularly inherent informant accuracy assumptions, are necessary for SNA analysts to ascertain the resultant social network model's limitations and the inferences that can properly be drawn from the analysis. Specific to investigating terrorist groups and dark networks, trusted and deceptive social network information sources are introduced.

PREDICTORS OF SEVERITY AND DURATION OF INFECTIOUS DISEASE OUTBREAKS IN LONG-TERM CARE FACILITIES IN THE SASKATOON HEALTH REGION

IntroductionInfectious disease outbreaks in long-term care facilities (LTCFs) pose a significant threat to the health and well-being of residents. While prevention of all outbreaks is not realistic, minimising the severity...

A Phylogeny of Birds Based on Over 1,500 Loci Collected by Target Enrichment and High-Throughput Sequencing

Evolutionary relationships among birds in Neoaves, the clade comprising the vast majority of avian diversity, have vexed systematists due to the ancient, rapid radiation of numerous lineages. We applied a new phylogenomic approach to resolve relationships in Neoaves using target enrichment (sequence capture) and high-throughput sequencing of ultraconserved elements (UCEs) in avian genomes. We collected sequence data from UCE loci for 32 members of Neoaves and one outgroup (chicken) and analyzed data sets that differed in their amount of missing data. An alignment of 1,541 loci that allowed missing data was 87% complete and resulted in a highly resolved phylogeny with broad agreement between the Bayesian and maximum-likelihood (ML) trees. Although results from the 100% complete matrix of 416 UCE loci were similar, the Bayesian and ML trees differed to a greater extent in this analysis, suggesting that increasing from 416 to 1,541 loci led to increased stability and resolution of the tree. Novel results of our study include surprisingly close relationships between phenotypically divergent bird families, such as tropicbirds (Phaethontidae) and the sunbittern (Eurypygidae) as well as between bustards (Otididae) and turacos (Musophagidae). This phylogeny bolsters support for monophyletic waterbird and landbird clades and also strongly supports controversial results from previous studies, including the sister relationship between passerines and parrots and the non-monophyly of raptorial birds in the hawk and falcon families. Although significant challenges remain to fully resolving some of the deep relationships in Neoaves, especially among lineages outside the waterbirds and landbirds, this study suggests that increased data will yield an increasingly resolved avian phylogeny.

Wavelet frame based color image demosaicing

Color image demosaicing consists in recovering full resolution color information from color-filter-array (CFA) samples with 66.7% amount of missing data. Most of the existing color demosaicing methods [14, 25, 16, 2, 26] are based on interpolation from inter-channel correlation and local geometry, which are not robust to highly saturated color images with small geometric features. In this paper, we introduce wavelet frame based methods by using a sparse wavelet [8, 22, 9, 23] approximation of individual color channels and color differences that recovers both geometric features and color information. The proposed models can be efficiently solved by Bregmanized operator splitting algorithm [27]. Numerical simulations of two datasets: McM and Kodak PhotoCD, show that our method outperforms other existing methods in terms of PSNR and visual quality.

Fast error analysis of continuous GNSS observations with missing data

One of the most widely used method for the time-series analysis of continuous Global Navigation Satellite System (GNSS) observations is Maximum Likelihood Estimation (MLE) which in most implementations requires $$\mathcal{O }(n^3)$$ operations for $$n$$ observations. Previous research by the authors has shown that this amount of operations can be reduced to $$\mathcal{O }(n^2)$$ for observations without missing data. In the current research we present a reformulation of the equations that preserves this low amount of operations, even in the common situation of having some missing data.Our reformulation assumes that the noise is stationary to ensure a Toeplitz covariance matrix. However, most GNSS time-series exhibit power-law noise which is weakly non-stationary. To overcome this problem, we present a Toeplitz covariance matrix that provides an approximation for power-law noise that is accurate for most GNSS time-series.Numerical results are given for a set of synthetic data and a set of International GNSS Service (IGS) stations, demonstrating a reduction in computation time of a factor of 10–100 compared to the standard MLE method, depending on the length of the time-series and the amount of missing data.

Applying time series models with spatial correlation to identify the scale of variation in habitat metrics related to threatened coho salmon (Oncorhynchus kisutch) in the Pacific Northwest

Trend analyses are common in the analysis of fisheries data, yet the majority of them ignore either observation error or spatial correlation. In this analysis, we applied a novel hierarchical Bayesian state-space time series model with spatial correlation to a 12-year data set of habitat variables related to coho salmon ( Oncorhynchus kisutch ) in coastal Oregon, USA. This model allowed us to estimate the degree of spatial correlation separately for each habitat variable and the importance of observation error relative to environmental stochasticity. This framework allows us to identify variables that would benefit from additional sampling and variables where sampling could be reduced. Of the eight variables included in our analysis, we found three metrics related to habitat quality correlated at large spatial scales (gradient, fine sediment, shade cover). Variables with higher observation error (pools, active channel width, fine sediment) could be made more precise with more repeat visits. Our spatio-temporal model is flexible and extendable to virtually any spatially explicit monitoring data set, even with large amounts of missing data and no repeated observations. Potential extensions include fisheries catch data, abiotic indicators, invasive species, or species of conservation concern.

Major incident triage: Comparative validation using data from 7th July bombings

IntroductionThe importance of health service planning for major incident management has been recognised since the World Trade Centre attacks of September 2001 and is highly relevant to planning for the 2012 Olympics. UK national Major Incident guidance stipulates the use of a system of triage for casualties to prioritise treatment and ensure “the greatest good for the greatest number”. However, at least three triage systems are in use worldwide and no evidence exists to demonstrate their relative efficacy. The transport bombings in London on 7th July 2005 caused the largest number of casualties on mainland UK soil since World War 2. We aimed to validate three major incident triage systems using patient data from the 7th July bombings. Patients and methodsA retrospective cohort of patients from the 7th July bombings treated at the Royal London Hospital (RLH) was examined. Clinical information collected on arrival at RLH was used to allocate triage categories using the START, Manchester Sieve and CareFlight triage systems. The value of each system in identifying the critically injured patient was calculated. Results203 sets of records were examined. Outcome data was available for 166 patients, of whom 8 were critically injured. Of these 166 patients, triage categories could be retrospectively allocated for 124 (START), 127 (Manchester Sieve), 128 (CareFlight), including 4 of the critically injured. All three systems identified the same three patients as P1 or P2. The triage systems performed identically in identifying the critically injured, with sensitivity 50% and specificity 100% if using only the highest priority, or sensitivity 75% and specificity 99% if using the top 2 priority groups. Significant amounts of data were not recorded in prehospital and hospital notes. Discussion and conclusionsSystematic triage of mass casualties is effective but the amount of missing data seriously compromises any attempt to evaluate systems of trauma care in a major incident.

A review of the reporting and handling of missing data in cohort studies with repeated assessment of exposure measures

BackgroundRetaining participants in cohort studies with multiple follow-up waves is difficult. Commonly, researchers are faced with the problem of missing data, which may introduce biased results as well as a loss of statistical power and precision. The STROBE guidelines von Elm et al. (Lancet, 370:1453-1457, 2007); Vandenbroucke et al. (PLoS Med, 4:e297, 2007) and the guidelines proposed by Sterne et al. (BMJ, 338:b2393, 2009) recommend that cohort studies report on the amount of missing data, the reasons for non-participation and non-response, and the method used to handle missing data in the analyses. We have conducted a review of publications from cohort studies in order to document the reporting of missing data for exposure measures and to describe the statistical methods used to account for the missing data.MethodsA systematic search of English language papers published from January 2000 to December 2009 was carried out in PubMed. Prospective cohort studies with a sample size greater than 1,000 that analysed data using repeated measures of exposure were included.ResultsAmong the 82 papers meeting the inclusion criteria, only 35 (43%) reported the amount of missing data according to the suggested guidelines. Sixty-eight papers (83%) described how they dealt with missing data in the analysis. Most of the papers excluded participants with missing data and performed a complete-case analysis (n = 54, 66%). Other papers used more sophisticated methods including multiple imputation (n = 5) or fully Bayesian modeling (n = 1). Methods known to produce biased results were also used, for example, Last Observation Carried Forward (n = 7), the missing indicator method (n = 1), and mean value substitution (n = 3). For the remaining 14 papers, the method used to handle missing data in the analysis was not stated.ConclusionsThis review highlights the inconsistent reporting of missing data in cohort studies and the continuing use of inappropriate methods to handle missing data in the analysis. Epidemiological journals should invoke the STROBE guidelines as a framework for authors so that the amount of missing data and how this was accounted for in the analysis is transparent in the reporting of cohort studies.

On Estimating Wet Canopy Evaporation from Deciduous and Coniferous Forests in the Asian Monsoon Climate

Abstract Continuous and direct measurement of evapotranspiration (ET) by the eddy covariance (EC) technique is still a challenge under monsoon climate because of a considerable amount of missing data during the long rainy periods and the consequential gap-filling process. Under such wet canopy conditions, especially in forests, evaporation of the intercepted precipitation (EWC) contributes significantly to the total ET. To quantify the role of EWC, leaf wetness has been measured at multiple levels in the canopy simultaneously with eddy covariance measurements at the KoFlux Gwangneung deciduous and coniferous forests for the entire year from September 2007 to August 2008. In this study, the measured EWC and the controlling mechanism during the wet canopy conditions have been scrutinized. Based on the evaluation of the four different algorithms of EWC estimation, that of the variable infiltration capacity (VIC) land surface model (LSM) has been adopted. All the missing EWC data are then recalculated by using the algorithm of VIC LSM and compared against the traditionally gap-filled EWC data based on the modified lookup table (MLT) method. The latter consistently underestimated EWC on average by 39% in deciduous forest and by 28% in coniferous forest. Major causes of such differences were due to the failure of considering aerodynamic coupling, advection of sensible heat, and heat storage in the MLT-based gap-filling method. Accordingly, a new gap-filling strategy for EWC is proposed that takes proper controlling mechanisms into account.

Long‐term cognitive outcomes of infants born moderately and late preterm

To investigate whether infants born late preterm have poorer cognitive outcomes than term-born infants. A cohort study based on the Avon Longitudinal Study of Parents and Children. Cognitive measures were assessed between the ages of 8 and 11 years. Exposure groups were defined as moderate/late preterm (32-36 weeks' gestation) or term (37-42 wk). Regression models were used to investigate the association between gestational age and IQ. Seven hundred and forty-one infants (5.4% of total eligible population; 422 males, 319 females; mean (SD) birthweight 2495 g [489]) were born between 32 and 36 weeks' gestation. The analysis was based on 6957 infants with IQ data at age 11 (50% of eligible infants). In the adjusted model, children born moderately and late preterm had similar IQ scores to peers born at term (mean difference [95% confidence interval] -0.18 [-1.88 to 1.52]). However, the preterm infants had a higher risk of having special educational needs at school (odds ratio 1.56 [1.18-2.07]). Despite an increased risk of special educational needs, there is little evidence of a reduction in IQ, memory, or attention measures at school age in children born between 32 and 36 weeks' gestation. Although interpretation is limited by the amount of missing data, further work is needed to identify why these infants have increased educational needs.

Associations between pre-operative radiographic changes and outcomes after total knee joint replacement for osteoarthritis

Purpose: Knee joint replacement (KJR) for OA is a successful intervention, but about 20% of KJR patients obtain little or no improvement. The reasons for this are unclear. The purpose of this study was to investigate whether pre-operative radiographic changes predict pain and functional outcomes. Methods: Data from a prospective single-centre cohort study of patients undergoing TJR between Jan 1st 2006 and Dec 31st 2007 were analysed (n = 525). Pre-operative data collection included demographics (age, sex, BMI, self-reported co-morbidities), the International Knee Society Score (IKSS), and the Short Form Health Survey (SF-12). Pre-operative (within 6 months of surgery) AP standing and lateral radiographs were read by a single observer using the Kellgren and Laurence (K&L) and Altman atlases, and the scoring repeated on a random 10% sample to assess rater reliability. Post-operative data collection at 12 and 24 months included the IKSS and SF-12. Linear and logistic regression analyses were undertaken to assess the relationships between baseline radiographic features and pre and post-operative pain and function. Results: Of the 525 cases 24 were excluded as they had undergone uni-compartmental replacements, and 23 radiographs were rejected, leaving 478 TJR cases for inclusion. 5 and 24 were unable to complete the 12 or 24 month data respectively due to death, loss to follow-up or refusal. The amount of missing data was well below 5%. The mean age of the subjects was 70.8 years (+/- 8.3), 69% were female and the mean BMI was 32.2. Pain scores improved from a mean of 4.0(+/-7.3) to 34.9 at 12 months and 34.8 (+/- 15.9) at 24 months; however about 30% of patients still complained of moderate or severe pain in the operated knee at each time point. Function scores improved from a mean of 37.5 (+/- 18.1) pre-operatively to 58.8 and 55.5 (+/- 26.8) at one and two years respectively. Inter-rater reliability scores for radiographic features were satisfactory, ranging from a kappa of 0.80 (lateral joint space narrowing) to 0.65 (lateral osteophyte score). The majority had OA K&L grade 3 (57 with mild (3a) and 200 with severe (3b) JSN) or 4 (87 without (4a) and 119 with (4b) bone attrition). The commonest compartmental distributions were medial and patella-femoral (244), or medial only (104). There were no significant associations between pre-operative radiographic scores and pre-operative pain or function, but radiographic severity was associated with outcome. Those with lower K&L grades and less compartmental involvement were less likely to have obtained improvements in pain at either 12 or 24 months (e.g. the odds ratio for ongoing pain in the operated joint at 12 months comparing those with K&L grades <3 to those with grade 4b was 5.23 (CI 1.5 to 17.6), p=0.008). The relationships between pre-operative radiographs and function were similar but less strong. Outcome did not appear to differ for those with predominant medial, lateral or patella-femoral joint involvement. Multivariate analyses showed that other significant, independent determinants of improvement in pain and function included age, mental function score of the SF-12 and pre-operative pain and function (both pain and function), as well as gender and BMI (function only). Conclusions: The data suggest that we should exercise caution when considering surgery for patients with severe pain but mild radiographic changes, and conversely, that those with severe joint damage, irrespective of which compartments are involved, are more likely to respond well to surgery. A possible explanation for this is that in those with severe pain and mild joint damage the pain is driven more by central sensitisation than by any nociceptive input from the joint.

Pain control and delay in time to skeletal-related events (SREs) in patients with metastatic castration-resistant prostate cancer (mCRPC) treated with abiraterone acetate (AA): Long-term follow-up.

183 Background: In study COU-AA-301, the androgen biosynthesis inhibitor AA significantly increased overall survival in mCRPC post-docetaxel (D) in both interim and updated analyses at 552 and 775 events, respectively. Interim data also showed a significant benefit of AA on patient (pt) reported pain. We present long-term outcomes of the effect of AA on pain and SREs. Methods: In this international randomized double blind study of AA (1 g QD) + prednisone (P) (5 mg BID) vs placebo + P in mCRPC post-D, pain was assessed at baseline and each treatment cycle until treatment discontinuation using the BPI-SF questionnaire. Palliation/progression of pain intensity (P-INT) and pain interference (P-INF) were evaluated using a priori definitions. P-INT palliation was defined as ≥ 30% improvement in pt reported “worst pain in last 24 h” (on 0-10 numerical rating scale) durable for ≥ 28 d without increased analgesic use (by WHO criteria) in eligible pts (those with significant baseline pain). Conversely, P-INT progression was defined as ≥ 30% deterioration, or increased analgesic use, over ≥ 28 d. Results: 797 pts were randomized to AA + P, 398 to placebo + P. Median follow-up was 20.2 mos. Median time to SRE was 25.0 mos (AA + P) vs 20.3 mos (placebo + P), p = 0.0001. Pain data were available for most patients; the cumulative amount of missing data ranged from 5% at Cycle 1 to 7% at Cycle 20. AA + P improved P-INT outcomes (Table); P-INF benefits were similar and also highly significant. Conclusions: Benefits of AA + P observed in the interim analysis, including superior and more durable pain relief and delay of pain progression and SREs, were maintained with longer follow-up. Additional analyses of the potential relationships between pain/SREs and other outcomes, as well as of prednisone’s effect on pain palliation in the control arm, are ongoing. [Table: see text]

Reports of Increased Mortality and GH: Will This Affect Current Clinical Practice?

Although randomized controlled studies are considered the “gold standard” to assess efficacy and safety, this approach is not always practical to ascertain long-term safety. To better define emerging safety signals, particularly for products already on the market, we mostly rely on observational studies. However, we are cognizant that all observational studies suffer from a critical deficiency: the design is not an experimental one. Because each patient's treatment is deliberately chosen rather than randomly assigned, the risk of selection bias is unavoidable. As a result, systematic differences in outcomes may not be due to the treatment itself. Although methods to adjust for identifiable differences are available, it is impossible to be certain that such adjustments are sufficient or whether they address all patients' relevant characteristics. Estimates of the magnitude of the treatment effect and on the generalizability of the findings are therefore imprecise. All these caveats apply to any observational study. Observational studies do not definitively answer a question, but rather generate hypotheses that need to be further explored with additional studies. Questions regarding the short- and long-term safety of GH have been gaining attention due to the expansion of GH indications and the use of increased dosages. In this issue of the JCEM, Carel et al. (1) report the results of a longitudinal follow-up study of subjects receiving GH where they ascertained its long-term effects on mortality. By 2009, after a mean follow-up of 17.3 yr, they were able to determine the vital status of about 95% of approximately 7000 subjects who received GH in France. Indications for GH varied, but 75% were labeled as idiopathic GH deficient (IGHD), 11.5% as idiopathic short stature, 8% as having a GH neurosecretory disorder, and 5.5% as small for gestational age. Ninety-three subjects died. Eighty-six deaths occurred after GH discontinuation, but six occurred while on active treatment. All-cause mortality was 33% higher than expected when compared with the general population. This increase was attributed to a higher than expected number of bone tumors, as well as subarachnoid and intracerebral hemorrhages. The increases were not associated with any other types of cancer. Doses in excess of 50 μg/kg·d were strongly associated with the increase in mortality. Significance was seen only after 15 yr of follow-up. Analyses at 5 and 10 yr did not show a significant increase in mortality compared with the general population. Shorter children at baseline and males were at increased risk, as well as those with more robust responses to provocative tests. The strengths of this study (1) include the balanced review and fair presentation of the information; the extent of the data, including diagnoses, dosages, time of exposure, and long-term follow-up; the attempts to ascertain in depth causes of death; and the use of internal and external controls. A major weakness of this study (1) is the lack of efficacy data. We do not know how these subjects responded to the treatment. Other limitations are related to the inability to identify and secure an appropriate short stature comparator, the lack of data on growth velocity before GH was administered, the absence of IGF-I levels before and during GH, the small number of events, and the significant amount of missing data, particularly in the ascertainment of cause of death in 22% of cases. The mean age at the time of censoring was less than 30 yr, and it is unknown whether the encountered trend will continue or disappear with passage of time. The number of females may be too small to detect any effect in this group. Many of these shortcomings were recognized by the authors, and attempts to address them were made. Each of these deficiencies, however, opens the door to alternative explanations. It is important to stress that the difficulties facing these authors are common in the field, and that investigators undertaking this sort of project are likely to encounter similar barriers.

Impact of missing data, gene choice, and taxon sampling on phylogenetic reconstruction: the Caryophyllales (angiosperms)

Density of taxon sampling and number/kind of characters are central to achieving the ultimate goals in phylogenetic reconstruction: tree robustness and improved accuracy. In molecular phylogenetics, DNA sequence repositories such as GenBank are potential sources for expanding datasets in two dimensions, taxa and characters, to the level of “supermatrices.” However, the issue of missing characters/genomic regions is generally considered a major impediment to this endeavor. We used here the angiosperm order Caryophyllales to systematically address the impact of missing data when expanding taxon sampling and number of characters in phylogenetic reconstruction. Our analyses show that expansion of taxon sampling by ~13-fold resulted in improved phylogenetic assessment of the Caryophyllales despite up to 38% missing data. Expanding number of characters in the dataset by allowing for up to 100-fold increase in amount of missing data and inclusion of entries with about 40% missing genomic regions did not negatively impact tree structure or robustness, but to the contrary improved both. These results are timely regarding the ongoing efforts to achieve detailed assessment of the tree of life.

The Impact of Missing Data on the Detection of Nonuniform Differential Item Functioning

Missing information is a ubiquitous aspect of data analysis, including responses to items on cognitive and affective instruments. Although the broader statistical literature describes missing data methods, relatively little work has focused on this issue in the context of differential item functioning (DIF) detection. Such prior research has focused on uniform DIF and has identified a complicated relationship between the type and amount of missing data, along with the methods for dealing with it. The current simulation study was designed to extend this earlier work by focusing on the impact of missing data and methods for dealing with it on nonuniform DIF detection using popular tools, including logistic regression, crossing SIBTEST, and the item response theory likelihood ratio test. Results suggest that the relationships between the various factors manipulated here were complex with no one method emerging as optimal in all cases, although listwise deletion consistently produced results similar to those obtained with the complete data across simulated conditions. Implications of these results are also discussed.