Variable Selection Techniques Research Articles

SME default prediction: A systematic methodology-focused review

ABSTRACT This study reviews the methodologies used in the literature to predict failure in small and medium-sized enterprises (SMEs). We identified 145 SMEs’ default prediction studies from 1972 to early 2023. We summarized the methods used in each study. The focus points are estimation methods, sample re-balancing methods, variable selection techniques, validation methods, and variables included in the literature. More than 1,200 factors used in failure prediction models have been identified, along with 54 unique feature selection techniques and 80 unique estimation methods. Over one-third of the studies do not use any feature selection method, and more than one-quarter use only in-sample validation. Our main recommendation for researchers is to use feature selection and validate results using hold-out samples or cross-validation. As an avenue for further research, we suggest in-depth empirical comparisons of estimation methods, feature selection techniques, and sample re-balancing methods based on some large and commonly used datasets.

Mini Percutaneous Nephrolithotomy vs Standard Percutaneous Nephrolithotomy: A Perioperative Decision Support System for Surgical Success Comparison.

This study aimed to rank the features that are important in terms of safety and effectiveness in choosing the surgical method and providing appropriate care to the patient by using the variables examined before and after the surgery to evaluate the success of mini percutaneous nephrolithotomy and standard percutaneous nephrolithotomy surgeries. The features evaluated before and after surgery were ranked according to their importance in the features considered, using Multivariate Adaptive Regression Splines (MARS), LASSO, Ridge, Elastic_net, and Random Forest algorithms as variable selection techniques. There are 278 samples in the relevant data set. Type of surgery (100%), intercostal access (97.75%), kidney opening procedure (94.25%), postoperative creatinine (59.22%), hydronephrosis (52.23%), the number of entries (41.61%), and pre- and post-operative hemoglobin difference (45.13%) were determined as the most critical variables. The MARS algorithm showed the most successful performance, with the lowest mean absolute error (MAE) value of 0.3622, the lowest root mean square error (RMSE) value of 0.3960, and the highest R2 value of 0.3405. Clinical decision support systems can be helpful in eliminating errors and reducing costs. It can also improve the quality of healthcare and aid in the early diagnosis of diseases. Computer-aided decision-making systems can be developed using the results of such products. These systems can provide doctors with better information about their patient's treatment options and improve decision-making. It can contribute to patients being better informed about the surgery results and taking an active role. In conclusion, this study provides essential information that should be included in the surgical decision-making process for patients using medications and with a history of percutaneous nephrolithotomy.

Proposed Two-Steps Procedure of Classification High Dimensional Data with Regularized Logistic Regression

The field of Bioinformatics has developed in response to the rapid increase in biological data, particularly high-dimensional gene expression data. Bioinformatics utilizes optimization, computational science, and statistical methods to effectively address challenges in the field of molecular biology. Numerous genes (variables) in gene expression are irrelevant to their study. Gene selection has been demonstrated to be an effective means of enhancing the performance of numerous methods of classification. The job of acquiring significant variables via the use of ranking variable selection (RVS) techniques and then picking the most effective classifier is an enormous challenge in the context of high-dimensional data. in this study, we proposed a new ranking filter method using smooth clipped absolute deviation depending on the resampling technique (RSVS) to obtain a proficient subset of genes with strong classification abilities. This is achieved by merging A screening technique employed as a filtering method in conjunction with Regularized Logistic Regression, such as LASSO,ALASSO,ENET, and MCP. The study involved the utilization of both simulated and real datasets to conduct an empirical evaluation of the proposed approach. The findings indicated that the proposed method outperformed other established methods. it was tested using three publicly data sets about Cancer. The Results demonstrate that the suggested approach is highly effective and viable, thus showing a strong level of performance with regards to accuracy, geometric mean, and the area under the curve. Furthermore, The findings suggest that the genes most often chosen are physiologically associated with the specific form of cancer. Therefore, the method that has been suggested has potential advantages for the classification of cancer via the use of DNA gene expression data within a clinical setting.

Variable Selection for Length-Biased and Interval-Censored Failure Time Data

Length-biased failure time data occur often in various biomedical fields, including clinical trials, epidemiological cohort studies and genome-wide association studies, and their analyses have been attracting a surge of interest. In practical applications, because one may collect a large number of candidate covariates for the failure event of interest, variable selection becomes a useful tool to identify the important risk factors and enhance the estimation accuracy. In this paper, we consider Cox’s proportional hazards model and develop a penalized variable selection technique with various popular penalty functions for length-biased data, in which the failure event of interest suffers from interval censoring. Specifically, a computationally stable and reliable penalized expectation-maximization algorithm via two-stage data augmentation is developed to overcome the challenge in maximizing the intractable penalized likelihood. We establish the oracle property of the proposed method and present some simulation results, suggesting that the proposed method outperforms the traditional variable selection method based on the conditional likelihood. The proposed method is then applied to a set of real data arising from the Prostate, Lung, Colorectal and Ovarian cancer screening trial. The analysis results show that African Americans and having immediate family members with prostate cancer significantly increase the risk of developing prostate cancer, while having diabetes exhibited a significantly lower risk of developing prostate cancer.

SuSiE PCA: A scalable Bayesian variable selection technique for principal component analysis

Latent factor models, like principal component analysis (PCA), provide a statistical framework to infer low-rank representation in various biological contexts. However, feature selection is challenging when this low-rank structure manifests from a sparse subspace. We introduce SuSiE PCA, a scalable sparse latent factor approach that evaluates uncertainty in contributing variables through posterior inclusion probabilities. We validate our model in extensive simulations and demonstrate that SuSiE PCA outperforms other approaches in signal detection and model robustness. We apply SuSiE PCA to multi-tissue expression quantitative trait loci (eQTLs) data from GTEx v8 and identify tissue-specific factors and their contributing eGenes. We further investigate its performance on the large-scale perturbation data and find that SuSiE PCA identifies modules with a higher enrichment of ribosome-related genes than sparse PCA (false discovery rate [FDR] vs. ), while being 18x faster. Overall, SuSiE PCA provides an efficient tool to identify relevant features in high-dimensional biological data.

Forecasting DO of the river-type reservoirs using input variable selection and machine learning techniques - taking Shuikou reservoir in the Minjiang River as an example

Dissolved oxygen (DO) plays a significant role in maintaining the health of aquatic ecosystems. In this study, we propose a model that incorporates multiple machine learning methods for predicting DO. Firstly, the maximum information coefficient (MIC) was utilized to identify the key drivers of DO. Afterward, the particle swarm optimization (PSO) algorithm was used to enhance the traditional support vector regression (SVR) model, optimizing the penalty factor (c) and the width of the Gaussian kernel function (g). This resulted in the development of a MIC-PSO-SVR-based DO prediction model. As a case study, we analyzed three points (G1, G2, and Z1) located in the mainstem and tributaries of the Shuikou Reservoir in Fujian, China. The original dataset, encompassing various time scales and sample sizes, underwent reclassification. Key factors influencing DO were determined by calculating the MIC values between DO and each monitoring factor. The main findings of this study are as follows: (1) By assessing the correlation between candidate factors and DO, the MIC effectively eliminated irrelevant variables with low correlation, thereby reducing the dataset size. Furthermore, it was observed that the fluctuations in MIC values for each variable stabilized when the sample size exceeded 4000. (2) The model’s performance exhibited improvement with the reduced dataset. For instance, the mean absolute error (MAE) and root mean square error (RMSE) of the hybrid MIC-PSO-SVR model decreased by approximately 66% and 49%, respectively, whereas the R2 and Nash-Sutcliffe efficiency (NSE) are as high as 0.98 and 0.95, showing superior performance compared to the unreduced PSO-SVR model. (3) More importantly, the model successfully predicted sudden hypoxia events in the Shuikou reservoir area from October 11 to November 8, 2021. Additionally, the MIC-PSO-SVR model accurately captured DO changes in the point G2 in front of the Shuikou dam during a hypoxic event in the Shuikou reservoir. The prediction errors during hypoxia were as low as 0.23 mg·L-1 and 0.31 mg·L-1 for 1 h and 24 h, respectively.

Comparison between Variable-Selection Algorithms in PLS Regression with Near-Infrared Spectroscopy to Predict Selected Metals in Soil.

Soil is one of the Earth's most important natural resources. The presence of metals can decrease environmental quality if present in excessive amounts. Analyzing soil metal contents can be costly and time consuming, but near-infrared (NIR) spectroscopy coupled with chemometric tools can offer an alternative. The most important multivariate calibration method to predict concentrations or physical, chemical or physicochemical properties as a chemometric tool is partial least-squares (PLS) regression. However, a large number of irrelevant variables may cause problems of accuracy in the predictive chemometric models. Thus, stochastic variable-selection techniques, such as the Firefly algorithm by intervals in PLS (FFiPLS), can provide better solutions for specific problems. This study aimed to evaluate the performance of FFiPLS against deterministic PLS algorithms for the prediction of metals in river basin soils. The samples had their spectra collected from the region of 1000-2500 nm. Predictive models were then built from the spectral data, including PLS, interval-PLS (iPLS), successive projections algorithm for interval selection in PLS (iSPA-PLS), and FFiPLS. The chemometric models were built with raw data and preprocessed data by using different methods such as multiplicative scatter correction (MSC), standard normal variate (SNV), mean centering, adjustment of baseline and smoothing by the Savitzky-Golay method. The elliptical joint confidence region (EJCR) used in each chemometric model presented adequate fit. FFiPLS models of iron and titanium obtained a relative prediction deviation (RPD) of more than 2. The chemometric models for determination of aluminum obtained an RPD of more than 2 in the preprocessed data with SNV, MSC and baseline (offset + linear) and with raw data. The metals Be, Gd and Y failed to obtain adequate models in terms of residual prediction deviation (RPD). These results are associated with the low values of metals in the samples. Considering the complexity of the samples, the relative error of prediction (REP) obtained between 10 and 25% of the values adequate for this type of sample. Root mean square error of calibration and prediction (RMSEC and RMSEP, respectively) presented the same profile as the other quality parameters. The FFiPLS algorithm outperformed deterministic algorithms in the construction of models estimating the content of Al, Be, Gd and Y. This study produced chemometric models with variable selection able to determine metals in the Ipojuca River watershed soils using reflectance-mode NIR spectrometry.

Detection of influential observations in high-dimensional survival data

Survival analysis is a statistical technique mainly used to analyze time-to-event data. Identification of influential observation attains greater importance since it leads to discovering new prognostic factors. Influential observation in survival typically points to individuals whose survival time is extremely short or long in comparison to others. Particularly, when the data possess more covariates than the observations, all classical approaches fail to perform. Hence, dimensionality reduction is necessary for choosing appropriate variables and it has been done by popular techniques such as LASSO and elastic net algorithm. This paper consider high-dimensional breast cancer data, and its dimensionality is reduced using variable selection methods. Subsequently, the rank product test and martingale residuals are used to identify an influential observation. Furthermore, a resampling technique is used to validate the consistency and robustness of the methods. The novelty of this paper lies in comparing the prediction accuracy of datasets with and without outliers using Random Survival Forest (RSF) for different training fractions. Comparatively, the RSF result demonstrates that the LASSO approach outperform others in the absence of outliers. Therefore, we suggest reducing dimensionality using the LASSO variable selection technique first, followed by removing likely outliers to improve the performance of classification algorithms.

A Relaxation Approach to Feature Selection for Linear Mixed Effects Models

Linear Mixed-Effects (LME) models are a fundamental tool for modeling correlated data, including cohort studies, longitudinal data analysis, and meta-analysis. Design and analysis of variable selection methods for LMEs is more difficult than for linear regression because LME models are nonlinear. In this article we propose a novel optimization strategy that enables a wide range of variable selection methods for LMEs using both convex and nonconvex regularizers, including l 1 , Adaptive- l 1 , SCAD, and l 0 . The computational framework only requires the proximal operator for each regularizer to be readily computable, and the implementation is available in an open source python package pysr3, consistent with the sklearn standard. The numerical results on simulated data sets indicate that the proposed strategy improves on the state of the art for both accuracy and compute time. The variable selection techniques are also validated on a real example using a data set on bullying victimization. Supplementary materials for this article are available online.

Sensor Selection and State Estimation of Continuous mAb Production Processes

The production of monoclonal antibodies (mAbs) plays a pivotal role in therapeutic treatments, and optimizing their production is crucial for minimizing costs and improving their accessibility to patients. One way of improving the production process is to improve model accuracy through the correct estimation of its states and parameters. The contributions of this paper lie in the provision of guidelines for sensor selection in the upstream production process of mAbs to enhance the accuracy of state estimation. Furthermore, this paper applies an effective variable selection technique for simultaneous state and parameter estimations for enhanced estimation results in the biomanufacturing processes of mAbs. An estimation framework of MHE is designed for three different case studies to demonstrate the efficiency of the proposed approach. The estimation performance is compared and assessed using the Root Mean Squared Error (RMSE) as an evaluation criterion.

Inconsistency identification in network meta-analysis via stochastic search variable selection.

The reliability of the results of network meta-analysis (NMA) lies in the plausibility of the key assumption of transitivity. This assumption implies that the effect modifiers' distribution is similar across treatment comparisons. Transitivity is statistically manifested through the consistency assumption which suggests that direct and indirect evidence are in agreement. Several methods have been suggested to evaluate consistency. A popular approach suggests adding inconsistency factors to the NMA model. We follow a different direction by describing each inconsistency factor with a candidate covariate whose choice relies on variable selection techniques. Our proposed method, stochastic search inconsistency factor selection (SSIFS), evaluates the consistency assumption both locally and globally, by applying the stochastic search variable selection method to determine whether the inconsistency factors should be included in the model. The posterior inclusion probability of each inconsistency factor quantifies how likely is a specific comparison to be inconsistent. We use posterior model odds or the median probability model to decide on the importance of inconsistency factors. Differences between direct and indirect evidence can be incorporated into the inconsistency detection process. A key point of our proposed approach is the construction of a reasonable "informative" prior concerning network consistency. The prior is based on the elicitation of information derived historical data from 201 published network meta-analyses. The performance of our proposed method is evaluated in two published network meta-analyses. The proposed methodology is publicly available in an R package called ssifs, published on CRAN and developed and maintained by the authors of this work.

A comparative study of different variable selection methods based on numerical simulation and empirical analysis.

This study employs the principles of computer science and statistics to evaluate the efficacy of the linear random effect model, utilizing Lasso variable selection techniques (including Lasso, Elastic-Net, Adaptive-Lasso, and SCAD) through numerical simulation and empirical research. The analysis focuses on the model's consistency in variable selection, prediction accuracy, stability, and efficiency. This study employs a novel approach to assess the consistency of variable selection across models. Specifically, the angle between the actual coefficient vector β and the estimated coefficient vector is computed to determine the degree of consistency. Additionally, the boxplot tool of statistical analysis is utilized to visually represent the distribution of model prediction accuracy data and variable selection consistency. The comparative stability of each model is assessed based on the frequency of outliers. This study conducts comparative experiments of numerical simulation to evaluate a proposed model evaluation method against commonly used analysis methods. The results demonstrate the effectiveness and correctness of the proposed method, highlighting its ability to conveniently analyze the stability and efficiency of each fitting model.

Data fusion based on near-infrared spectroscopy and hyperspectral imaging technology for rapid adulteration detection of Ganoderma lucidum spore powder

As a function food resource with significant medicinal and economic value, Ganoderma lucidum spore powder (GLSP) is often adulterated with dyed starch by unscrupulous merchants in pursuit of higher profits. In this study, near-infrared spectroscopy (NIRS) and hyperspectral imaging techniques (HSI) were selected, and data fusion strategies were introduced to quickly detect the adulteration of GLSP. To enhance the model performance, 49 pretreatments and their combinations, as well as 5 feature variables selection techniques, were applied. The results showed that if only a single spectral technique was considered, NIRS was more advantageous than HSI in the adulteration identification and adulteration level prediction of GLSP. Mid-level fusion (MLF) strategy was more suitable for adulteration identification if data fusion strategy was introduced, with 100% accuracy, precision, recall and F1 score at random frog (RF) level. Therefore, NIRS and HSI data fusion can be used for quality evaluation of GLSP, which provides a new idea and potential method for the evaluation and detection of functional foods.

Identifying Prominent Environmental Covariates Using Variable Selection Methodologies for Digital Soil Mapping of Tamil Nadu, India

High dimensional datasets that depict intricate spatial variations are necessary to predict complex landscape structures and the corresponding soil properties taking into account the size of the research region in addition to the data attributes. The number and quality of the input datasets taken into consideration essentially determine the quantity and quality of the soil properties that may be predicted thanks to data-driven learning algorithms. The use of variable selection strategies both before and after the prediction can have a significant impact on the outcome and can lower the related computing load. The majority of commonly used variable selection techniques such as correlation analysis, stepwise regression and recursive feature elimination, among others perform recursive statistical/mathematical comparison to identify the significant covariates that improve the effectiveness of the algorithm proposed. In order to identify the effective environmental variables in predicting the soil attribute, this article investigated a widely used recursive ranking method called recursive feature elimination. The covariate layer that produced the lowest RMSE was placed first according to the rankings of the covariates provided by recursive feature elimination. The findings showed that among other factors physiography, mean rainfall, rock outcrop difference ratio, elevation and mean temperature will be effective in predicting the soil properties required for digital soil mapping.

Radiotherapy toxicity prediction using knowledge-constrained generalized linear model

Radiation therapy (RT) is a frontline approach to treating cancer. While the target of radiation dose delivery is the tumor, there is an inevitable spill of dose to nearby normal organs causing complications. This phenomenon is known as radiotherapy toxicity. To predict the outcome of the toxicity, statistical models can be built based on dosimetric variables received by the normal organ at risk (OAR), known as Normal Tissue Complication Probability (NTCP) models. To tackle the challenge of the high dimensionality of dosimetric variables and limited clinical sample sizes, statistical models with variable selection techniques are viable choices. However, existing variable selection techniques are data-driven and do not integrate medical domain knowledge into the model formulation. We propose a knowledge-constrained generalized linear model (KC-GLM). KC-GLM includes a new mathematical formulation to translate three pieces of domain knowledge into non-negativity, monotonicity, and adjacent similarity constraints on the model coefficients. We further propose an equivalent transformation of the KC-GLM formulation, which makes it possible to solve the model coefficients using existing optimization solvers. Furthermore, we compare KC-GLM and several well-known variable selection techniques via a simulation study and on two real datasets of prostate cancer and lung cancer, respectively. These experiments show that KC-GLM selects variables with better interpretability, avoids producing counter-intuitive and misleading results, and has better prediction accuracy.

Comparing Factors Associated with Increased Stimulant Use in Relation to HIV Status Using a Machine Learning and Prediction Modeling Approach.

Stimulant use is an important driver of HIV/STI transmission among men who have sex with men (MSM). Evaluating factors associated with increased stimulant use is critical to inform HIV prevention programming efforts. This study seeks to use machine learning variable selection techniques to determine characteristics associated with increased stimulant use and whether these factors differ by HIV status. Data from a longitudinal cohort of predominantly Black/Latinx MSM in Los Angeles, CA was used. Every 6months from 8/2014-12/2020, participants underwent STI testing and completed surveys evaluating the following: demographics, substance use, sexual risk behaviors, and last partnership characteristics. Least absolute shrinkage and selection operator (lasso) was used to select variables and create predictive models for an interval increase in self-reported stimulant use across study visits. Mixed-effects logistic regression was then used to describe associations between selected variables and the same outcome. Models were also stratified based on HIV status to evaluate differences in predictors associated with increased stimulant use. Among 2095 study visits from 467 MSM, increased stimulant use was reported at 20.9% (n = 438) visits. Increased stimulant use was positively associated with unstable housing (adjusted [a]OR 1.81; 95% CI 1.27-2.57), STI diagnosis (1.59; 1.14-2.21), transactional sex (2.30; 1.60-3.30), and last partner stimulant use (2.21; 1.62-3.00). Among MSM living with HIV, increased stimulant use was associated with binge drinking, vaping/cigarette use (aOR 1.99; 95% CI 1.36-2.92), and regular use of poppers (2.28; 1.38-3.76). Among HIV-negative MSM, increased stimulant use was associated with participating in group sex while intoxicated (aOR 1.81; 95% CI 1.04-3.18), transactional sex (2.53; 1.40-2.55), and last partner injection drug use (1.96; 1.02-3.74). Our findings demonstrate that lasso can be a useful tool for variable selection and creation of predictive models. These results indicate that risk behaviors associated with increased stimulant use may differ based on HIV status and suggest that co-substance use and partnership contexts should be considered in the development of HIV prevention/treatment interventions.

Simultaneous selection of multiple important single nucleotide polymorphisms in familial genome wide association studies data

We propose a resampling-based fast variable selection technique for detecting relevant single nucleotide polymorphisms (SNP) in a multi-marker mixed effect model. Due to computational complexity, current practice primarily involves testing the effect of one SNP at a time, commonly termed as ‘single SNP association analysis’. Joint modeling of genetic variants within a gene or pathway may have better power to detect associated genetic variants, especially the ones with weak effects. In this paper, we propose a computationally efficient model selection approach—based on the e-values framework—for single SNP detection in families while utilizing information on multiple SNPs simultaneously. To overcome computational bottleneck of traditional model selection methods, our method trains one single model, and utilizes a fast and scalable bootstrap procedure. We illustrate through numerical studies that our proposed method is more effective in detecting SNPs associated with a trait than either single-marker analysis using family data or model selection methods that ignore the familial dependency structure. Further, we perform gene-level analysis in Minnesota Center for Twin and Family Research (MCTFR) dataset using our method to detect several SNPs using this that have been implicated to be associated with alcohol consumption.

Combining phenotypic and genomic data to improve prediction of binary traits

Plant breeders want to develop cultivars that outperform existing genotypes. Some characteristics (here ‘main traits’) of these cultivars are categorical and difficult to measure directly. It is important to predict the main trait of newly developed genotypes accurately. In addition to marker data, breeding programs often have information on secondary traits (or ‘phenotypes’) that are easy to measure. Our goal is to improve prediction of main traits with interpretable relations by combining the two data types using variable selection techniques. However, the genomic characteristics can overwhelm the set of secondary traits, so a standard technique may fail to select any phenotypic variables. We develop a new statistical technique that ensures appropriate representation from both the secondary traits and the genotypic variables for optimal prediction. When two data types (markers and secondary traits) are available, we achieve improved prediction of a binary trait by two steps that are designed to ensure that a significant intrinsic effect of a phenotype is incorporated in the relation before accounting for extra effects of genotypes. First, we sparsely regress the secondary traits on the markers and replace the secondary traits by their residuals to obtain the effects of phenotypic variables as adjusted by the genotypic variables. Then, we develop a sparse logistic classifier using the markers and residuals so that the adjusted phenotypes may be selected first to avoid being overwhelmed by the genotypic variables due to their numerical advantage. This classifier uses forward selection aided by a penalty term and can be computed effectively by a technique called the one-pass method. It compares favorably with other classifiers on simulated and real data.

Association of Early Serum Phosphate Levels and Mortality in Patients with Sepsis.

Metabolic derangements in sepsis influence phosphate levels, which may predict mortality outcomes. We investigated the association between initial phosphate levels and 28-day mortality in patients with sepsis. We conducted a retrospective analysis of patients with sepsis. Initial (first 24 hours) phosphate levels were divided into phosphate quartile groups for comparisons. We used repeated-measures mixed-models to assess differences in 28-day mortality across the phosphate groups, adjusting for other predictors identified by the Least Absolute Shrinkage and Selection Operator variable selection technique. A total of 1,855 patients were included with 13% overall 28-day mortality (n=237). The highest phosphate quartile (>4.0 milligrams per deciliter [mg/dL]) had a higher mortality rate (28%) than the three lower quartiles (P<0.001). After adjustment (age, organ failure, vasopressor administration, liver disease), the highest initial phosphate was associated with increased odds of 28-day mortality. Patients in the highest phosphate quartile had 2.4 times higher odds of death than the lowest (≤2.6 mg/dL) quartile (P<0.01), 2.6 times higher than the second (2.6-3.2 mg/dL) quartile (P<0.01), and 2.0 times higher than the third (3.2-4.0 mg/dL) quartile (P=0.04). Septic patients with the highest phosphate levels had increased odds of mortality. Hyperphosphatemia may be an early indicator of disease severity and risk of adverse outcomes from sepsis.

Are systemic risk measures effective? Evidence from macroeconomic downside risk prediction

ABSTRACT This paper studies the predictive power of a set of systemic risk measures for macroeconomic downside risks in China. The quantile single-index model (QSIM) is used to select and combine systemic risk measures to depict the lower tail of macroeconomy distribution. The out-of-sample results show that QSIM provides prominent and robust prediction, as it aggregates all relevant systemic risk measures information and captures parameter nonlinearity. Moreover, we uncover the risk drivers behind macroeconomic downturns using the variable selection technique. The results provide an early warning mechanism for regulators and a criterion for researchers to select systemic risk measures.