Differences In Predictive Ability Research Articles

Child Mortality as Predicted by Nutritional Status and Recent Weight Velocity in Children under Two in Rural Africa

AbstractWHO has released prescriptive child growth standards for, among others, BMI-for-age (BMI-FA), mid-upper arm circumference-for-age, and weight velocity. The ability of these indices to predict child mortality remains understudied, although growth velocity prognostic value underlies current growth monitoring programs. The study aims were first to assess, in children under 2, the independent and combined ability of these indices and of stunting to predict all-cause mortality within 3 mo, and second, the comparative abilities of weight-for-length (WFL) and BMI-FA to predict short-term (<3 mo) mortality. We used anthropometry and survival data from 2402 children aged between 0 and 24 mo in a rural area of the Democratic Republic of Congo with high malnutrition and mortality rates and limited nutritional rehabilitation. Analyses used Cox proportional hazard models and receiver operating characteristic curves. Univariate analysis and age-adjusted analysis showed predictive ability of all indices. Multivariate analysis without age adjustment showed that only very low weight velocity [HR = 3.82 (95%CI = 1.91, 7.63);P< 0.001] was independently predictive. With age adjustment, very low weight velocity [HR = 3.61 (95%CI = 1.80, 7.25); P< 0.001] was again solely retained as an independent predictor. There was no evidence for a difference in predictive ability between WFL and BMI-FA. This paper shows the value of attained BMI-FA, a marker of wasting status, and recent weight velocity, a marker of the wasting process, in predicting child death using the WHO child growth standards. WFL and BMI-FA appear equivalent as predictors.

Comparison of multiple linear regression, partial least squares and artificial neural network for quantitative structure retention relationships of some polycyclic aromatic hydrocarbons

Quantitative structural-retention relationships (QSRR) of retention phenomena of polycyclic aromatic hydrocarbons (PAHs) facilitate resolving complex mixtures of them by gas chromatography (GC). The structural descriptors of 38 PAH compounds were calculated. Stepwise variable selection was applied for selection of meaningful descriptors. MLR, PLS and ANN models were built with calibration compounds. The predictive ability of the models was evaluated on 6 PAHs, which were not used in training steps and also by leave-6-out cross-validation. The best prediction results were obtained by the MLR model. The difference in predictive ability of the ANN and PLS model was trivial.

Combining static and dynamic variables in species distribution models under climate change

Summary1. Methods used to predict shifts in species’ ranges because of climate change commonly involve species distribution (niche) modelling using climatic variables, future values of which are predicted for the next several decades by general circulation models. However, species’ distributions also depend on factors other than climate, such as land cover, land use and soil type. Changes in some of these factors, such as soil type, occur over geologic time and are thus imperceptible over the timescale of these types of projections. Other factors, such as land use and land cover, are expected to change over shorter timescales, but reliable projections are not available. Some important predictor variables, therefore, must be treated as unchanging, or static, whether because of the properties of the variable or out of necessity. The question of how best to combine dynamic variables predicted by climate models with static variables is not trivial and has been dealt with differently in studies to date. Alternative methods include using the static variables as masks, including them as independent explanatory variables in the model, or excluding them altogether.2. Using a set of simulated species, we tested various methods for combining static variables with future climate scenarios. Our results showed that including static variables in the model with the dynamic variables performed better or no worse than either masking or excluding the static variables.3. The difference in predictive ability was most pronounced when there is an interaction between the static and dynamic variables.4. For variables such as land use, our results indicate that if such variables affect species distributions, including them in the model is better than excluding them, even though this may mean making the unrealistic assumption that the variable will not change in the future.5. These results demonstrate the importance of including static and dynamic non‐climate variables in addition to climate variables in species distribution models designed to predict future change in a species’ habitat or distribution as a result of climate change.

Is Timed Up and Go Better Than Gait Speed in Predicting Health, Function, and Falls in Older Adults?

To assess whether the Timed Up and Go (TUG) is superior to gait speed in predicting multiple geriatric outcomes. Prospective cohort study. Medicare health maintenance organization and Veterans Affairs primary care clinics. Adults aged 65 and older (N=457). Baseline gait speed and TUG were used to predict health decline according to EuroQol and Medical Outcomes Study 36-item Short Form Survey (SF-36) global health; functional decline according to National Health Interview Survey (NHIS) activities of daily living (ADLs) score and SF-36 physical function index; hospitalization; and any falls and multiple falls over 1 year. Mean age was 74, and 44% of participants were female. Odds ratios for all outcomes were equivalent for gait speed and TUG. Using area under the receiver operating characteristic curve of 0.7 or greater for acceptable predictive ability, gait speed and TUG each alone predicted decline in global health, new ADL difficulty, and falls, with no difference in predictive ability between performance measures. Neither performance measure predicted hospitalization, EuroQol decline, or physical function decline. As a continuous variable, TUG did not add predictive ability to gait speed for any outcome. Gait speed predicts most geriatric outcomes, including falls, as does TUG. The time taken to complete TUG may not add to information provided by gait speed, although its qualitative elements may have other utility.

Predictive Ability Assessment of Linear Mixed Models in Multienvironment Trials in Corn

ABSTRACTPrediction of future performance of cultivars is an important objective of multienvironment trials (MET). A series of linear mixed models with varying degrees of heterogeneous genotypic variance, correlation, and error variance structure were compared for their ability to predict performance in an untested environment in 51 data sets from the Iowa Crop Performance Test for corn (Zea mays L.). In most cases there was no substantial improvement in predictions among models that included heterogeneity of genotypic variance–covariance components, but the best prediction model included heterogeneous environment‐specific error variances in 63% of data sets analyzed. The largest differences in predictive ability among models appeared to be due to poor estimation of genotypic covariance components in data sets with few common hybrids across 2 yr in a data set. Simulation confirmed the observation from cross validation. Our results suggested that predictions were not improved by modeling heterogeneous genotypic covariance components because of the small number of common hybrids across years. Inclusion of heterogeneous error variances did lead to slight improvements in predictions.

Longitudinal changes in ultrasonic measurements of body composition during growth in Suffolk ram lambs and evaluation of alternative adjustment strategies for ultrasonic scan data

Four equations were used to compare alternative procedures to adjust ultrasonic estimates (y) of backfat thickness (BF) and LM area (LMA) for BW using data from a series of 7 scans on 24 Suffolk ram lambs born in 2007. Equations were linear, linear + quadratic, allometric (y = alphaBW(beta)), and allometric + BW (ABW; y = alphaBW(beta)e(gammaW)). Goodness of fit was very similar between equations over the range of the data. Resulting adjustment equations were tested using 3 serial scans on winter-born Suffolk (n = 150), Hampshire (n = 36), and Dorset (n = 43) rams and 52 fall-born Dorset rams tested at the Virginia Ram Test in 1999 through 2002. Partial correlations (accounting for the effect of year) between predicted and actual measures ranged from 0.78 to 0.87 for BF and 0.66 to 0.93 for LMA in winter-born rams and from 0.70 to 0.71 for BF and 0.72 to 0.78 for LMA in fall-born rams. No significant differences in predictive ability existed between equations for BF or LMA (P > 0.05), and there was no indication that the allometric equation was a better predictor than linear within the range of the data. Adjustment equations were also tested using serial scan data from 37 Suffolk ewe lambs born in the same contemporary group as the rams used to derive the prediction equations but fed for a substantially slower rate of BW gain. Correlations between predicted and actual values of BF and LMA indicated lambs were too young and small at the first scan (77 d, 32.4 kg) to reliably predict carcass measures at typical slaughter weights. For prediction using data from the 2 subsequent scans, at mean ages >96 d and mean BW >39 kg, correlations between predicted and actual values were 0.72 to 0.74 for BF and 0.54 to 0.76 for LMA. Little difference existed between equations for predicting BF. For LMA, the ABW form was a weaker predictor than the others, and the linear equation was slightly superior to allometric. Therefore, it appears the linear and allometric forms are both suitable for use in central ram test and performance-tested farm flocks.

Performance of several variable‐selection methods applied to real ecological data

I evaluated the predictive ability of statistical models obtained by applying seven methods of variable selection to 12 ecological and environmental data sets. Cross-validation, involving repeated splits of each data set into training and validation subsets, was used to obtain honest estimates of predictive ability that could be fairly compared among methods. There was surprisingly little difference in predictive ability among five methods based on multiple linear regression. Stepwise methods performed similarly to exhaustive algorithms for subset selection, and the choice of criterion for comparing models (Akaike's information criterion, Schwarz's Bayesian information criterion or F statistics) had little effect on predictive ability. For most of the data sets, two methods based on regression trees yielded models with substantially lower predictive ability. I argue that there is no 'best' method of variable selection and that any of the regression-based approaches discussed here is capable of yielding useful predictive models.

Predictive ability of models for calving difficulty in US Holsteins

The performance of alternative threshold models for analyzing calving difficulty (CD) in Holstein cows was evaluated in terms of predictive ability. Four models were considered, with CD classified into either three or four categories and analysed either as a single trait or jointly with gestation length (GL). The data contained GL and CD records from 90 393 primiparous cows, sired by 1122 bulls and distributed over 935 herd-calving year classes. Predictive ability of each model was evaluated using four criteria: mean squared error of the difference between observed and predicted CD scores; a Kullback-Leibler divergence measure between the observed and predicted distributions of CD scores; Pearson's correlation between observed and predicted CD scores and ability to correctly classify bulls as above or below average for incidence of CD. In general, the four models had similar predictive abilities. The joint analysis of CD with GL produced little, if any, improvement in predictive ability over univariate models. In light of the small difference in predictive ability between models treating CD with three or four categories and considering that a greater number of categories can provide more information, analysis of CD classified into four categories seems warranted.

Determination of specific gravity of green Pinus taeda samples by near infrared spectroscopy: comparison of pre-processing methods using multivariate figures of merit

Near infrared diffuse reflectance was used for the determination of specific gravity in green Pinus taeda L. wood samples representing simulated increment cores obtained at breast height and merchantable green logs. The effects of using three pre-processing methods (second derivative, multiplicative scatter correction, and orthogonal signal correction) to reduce the scatter observed in the original spectra were evaluated. The effectiveness of each method was assessed in terms of the average predictive ability of the models and in terms of multivariate figures of merit derived from net analyte signal theory. Specific gravity was successfully modeled using green wood samples. No differences in predictive ability among models were found, although more parsimonious regressions were obtained using transformed spectra. The incorporation of figures of merit for the characterization of calibration models proved to be a valuable tool for understanding the effects of the pre-processing alternatives on the final results.

Forecasting with DSGE Models: The Role of Nonlinearities

This paper studies whether the out-of-sample forecasting performance of a dynamic stochastic general equilibrium (DSGE) model improves by taking its nonlinear rather than its linear approximation to the data. We address this question within a New Keynesian monetary economy, considering both environments of simulated and real data. Precisely, we estimate our model based on its linear respectively quadratic approximate solution, generate out-of-sample forecasts for three observables (output, inflation, and the nominal interest rate), and compare the quality of forecasts by several statistical measures of accuracy. We find that the value of nonlinearities in terms of predictive power depends crucially on whether the model is well specified. For simulated data, the nonlinear model indeed forecasts noticeably better as compared to its linearized counterpart, whereas for real data, we find no substantial differences in predictive abilities.

A STATISTICAL COMPARISON OF MARYBLYT™ AND COUGARBLIGHT WITH SENSITIVITY ANALYSIS

MARYBLYT and Cougarblight are fire blight forecasters used to identify blossom infection periods for Erwinia amylovora on apple (Malus x domestica) and pear (Pyrus communis). MARYBLYT uses flowering, an index of bacterial population (EIP), a measurement of wetting, and average daily temperature, whereas Cougarblight uses flowering, wetting, a 4-day temperature window, and orchard fire blight history as parameters for predicting blossom infection. Sensitivity analysis was used to evaluate both models for their ability to predict fire blight using historical weather and disease incidence data collected from several regions. The results showed that the two forecasters performed similarly in their ability to predict blossom blight. However, regional differences in predictive ability were evident, with best performance of both forecasters occurring in England. Further analysis is needed to determine the reasons for the regional differences.

Comparative sediment quality guideline performance for predicting sediment toxicity in Southern California, USA

Several types of sediment quality guidelines (SQGs) are used by multiple agencies in southern California (USA) to interpret sediment chemistry data, yet little information is available to identify the best approaches to use. The objective of this study was to evaluate the predictive ability of five SQGs to predict the presence and absence of sediment toxicity in coastal southern California: the effects range-median quotient (ERMq), consensus moderate effect concentration (consensus MEC), mean sediment quality guideline quotient (SQGQ1), apparent effects threshold (AET), and equilibrium partitioning (EqP) for organics. Large differences in predictive ability among the SQGs were obtained when each approach was applied to the same southern California data set. Sediment quality guidelines that performed well in identifying nontoxic samples were not necessarily the best predictors of toxicity. In general, the mean ERMq, SQGQ1q, and consensus MECq approaches had a better overall predictive ability than the AET and EqP for organics approaches. In addition to evaluating the predictive ability of SQGs addressing chemical mixtures, the effect of an individual SQG value (DDT) was also evaluated for the mean ERMq with and without DDT. The mean ERMq without DDT had a better ability to predict toxic samples than the mean ERMq with DDT. Similarities in discriminatory ability between different approaches, variations in accuracy among SQG values for some chemicals, and the presence of complex mixtures of contaminants in most samples underscore the need to apply SQGs in combination, such as the mean quotient. Management objectives and SQG predictive ability using regional data should be determined beforehand so that the most appropriate SQG approach and critical values can be identified for specific applications.

A comparison of methods for testing differences in predictive ability

AbstractThis paper is devoted to the comparison of methods for testing difference in prediction ability of regression‐based predictors. Focus is on both parametric and non‐parametric methods. Comparisons are done using the bootstrap based on residuals from real data sets. The power of the methods for different alternatives is the main tool for evaluating the properties of the methods. The main conclusion from the bootstrap simulations is that the most suitable method seems to be two‐way analysis of variance of the absolute values of the prediction errors. Copyright © 2006 John Wiley &amp; Sons, Ltd.

Comparison between logistic regression and neural networks to predict death in patients with suspected sepsis in the emergency room

IntroductionNeural networks are new methodological tools based on nonlinear models. They appear to be better at prediction and classification in biological systems than do traditional strategies such as logistic regression. This paper provides a practical example that contrasts both approaches within the setting of suspected sepsis in the emergency room.MethodsThe study population comprised patients with suspected bacterial infection as their main diagnosis for admission to the emergency room at two University-based hospitals. Mortality within the first 28 days from admission was predicted using logistic regression with the following variables: age, immunosuppressive systemic disease, general systemic disease, Shock Index, temperature, respiratory rate, Glasgow Coma Scale score, leucocyte counts, platelet counts and creatinine. Also, with the same input and output variables, a probabilistic neural network was trained with an adaptive genetic algorithm. The network had three neurone layers: 10 neurones in the input layer, 368 in the hidden layer and two in the output layer. Calibration was measured using the Hosmer-Lemeshow goodness-of-fit test and discrimination was determined using receiver operating characteristic curves.ResultsA total of 533 patients were recruited and overall 28-day mortality was 19%. The factors chosen by logistic regression (with their score in parentheses) were as follows: immunosuppressive systemic disease or general systemic disease (2), respiratory rate 24–33 breaths/min (1), respiratory rate ≥ 34 breaths/min (3), Glasgow Come Scale score ≤12 (3), Shock Index ≥ 1.5 (2) and temperature <38°C (2). The network included all variables and there were no significant differences in predictive ability between the approaches. The areas under the receiver operating characteristic curves were 0.7517 and 0.8782 for the logistic model and the neural network, respectively (P = 0.037).ConclusionA predictive model would be an extremely useful tool in the setting of suspected sepsis in the emergency room. It could serve both as a guideline in medical decision-making and as a simple way to select or stratify patients in clinical research. Our proposed model and the specific development method – either logistic regression or neural networks – must be evaluated and validated in an independent population.

Performance-Based Evaluations of Guidelines for Nitrogen Fertilizer Application after Animal Manure

Nitrogen fertilizer needs for corn (Zea mays L.) in fields already treated with animal manure can be estimated by using general guidelines or soil testing for inorganic N. Although the soil-testing approach has been extensively evaluated for ability to predict yield responses to applied N under field conditions, the general-guideline approach has not been subjected to comparable performance-based evaluations. Fertilizer response trials were conducted in 205 manured fields to (i) compare the two approaches for ability to predict corn yield responses to fertilizer N applied after animal manure, (ii) identify reasons for differences in predictive ability, and (iii) explore the benefits of performance-based comparisons of the alternative approaches. Analyses showed that 34% of the observed variability in response could be explained by inorganic N concentrations whereas less than 5% of this variability could be explained by the general-guideline approach. The soil-testing approach, therefore, had greater ability to integrate the effects of all factors affecting yield responses across the range of conditions studied. Mean yield responses (0.55 Mg ha−1) were smaller than are usually detectable in individual trials, but they were great enough to prompt farmers to fertilize. Results of this study indicate that the most commonly accepted approach to estimating N fertilizer needs is less reliable than generally believed and, therefore, that superior approaches are likely to remain unrecognized unless the performance of the commonly accepted approach is objectively evaluated under realistic field conditions.

Performance‐Based Evaluations of Guidelines for Nitrogen Fertilizer Application after Animal Manure

Nitrogen fertilizer needs for corn (Zea mays L.) in fields already treated with animal manure can be estimated by using general guidelines or soil testing for inorganic N. Although the soil‐testing approach has been extensively evaluated for ability to predict yield responses to applied N under field conditions, the general‐guideline approach has not been subjected to comparable performance‐based evaluations. Fertilizer response trials were conducted in 205 manured fields to (i) compare the two approaches for ability to predict corn yield responses to fertilizer N applied after animal manure, (ii) identify reasons for differences in predictive ability, and (iii) explore the benefits of performance‐based comparisons of the alternative approaches. Analyses showed that 34% of the observed variability in response could be explained by inorganic N concentrations whereas less than 5% of this variability could be explained by the general‐guideline approach. The soil‐testing approach, therefore, had greater ability to integrate the effects of all factors affecting yield responses across the range of conditions studied. Mean yield responses (0.55 Mg ha−1) were smaller than are usually detectable in individual trials, but they were great enough to prompt farmers to fertilize. Results of this study indicate that the most commonly accepted approach to estimating N fertilizer needs is less reliable than generally believed and, therefore, that superior approaches are likely to remain unrecognized unless the performance of the commonly accepted approach is objectively evaluated under realistic field conditions.

An assessment of three creep–fatigue life prediction methods for nickel‐based superalloy GH4049

High‐temperature low‐cycle fatigue tests with and without a 10‐s strain hold period in a cycle were performed on a nickel base superalloy GH4049 under a fully reversed axial total strain control mode. Three creep–fatigue life prediction methods are chosen to analyse the experimental data. These methods are the linear damage summation method (LDS), the strain range partitioning method (SRP) and the strain energy partitioning method (SEP). Their ability to predict creep‐fatigue lives of GH4049 at 700, 800 and 850 °C has been evaluated. It is found that the SEP method shows an advantage over the SRP method for all the tests under consideration. At 850 °C, the LDS and SEP methods give a more satisfactory prediction for creep–fatigue lives. At the temperatures of 700 and 800 °C, the SRP and SEP methods can correlate the life data better than the LDS method. In addition, the differences in predictive ability of these methods have also been analysed. The scanning electron microscopy (SEM) examination of fracture surfaces reveals that under creep–fatigue test conditions crack initiation mode is transgranular, while crack propagation mode is either intergranular plus transgranular or entirely intergranular, dependent on test temperature.

The prognostic significance of troponin I and troponin T.

To determine and compare the prognostic abilities of early, single-sample measurements of cardiac troponin I (cTn-I), cardiac troponin T (cTn-T), and creatine kinase-MB (CK-MB) among ED patients with possible myocardial ischemia. Prospective collection of clinical and serologic data using an identity-unlinked technique from patients with possible myocardial ischemia at 2 urban EDs. Outcome data concerning the occurrence of adverse events (AEs) during the 14 days after enrollment were used to calculate and compare the relative risks (RRs) and predictive values (with 95% confidence intervals) of the 3 markers for predicting AEs. Among the 401 study patients, 105 AEs occurred in 67 patients. cTn-I, cTn-T, and CK-MB were all significantly predictive of AEs, with RRs of 3.87 (2.39, 6.26), 3.03 (1.92, 4.79), and 6.45 (4.74, 8.77), respectively. For prediction of AEs, sensitivity for each of the 3 markers was low (cTn-I = 15.38, cTn-T = 24.62, CK-MB = 15.38), while specificity was high (cTn-I = 97.62, cTn-T = 93.15, CK-MB = 99.70). No significant difference in predictive ability was found between cTn-I and cTn-T. However, a positive CK-MB result was a stronger predictor of AEs than either cTn-I (p = 0.01) or cTn-T (p = 0.001). No significant difference in predictive abilities was found between cTn-I and cTn-T. However, routine testing for both CK-MB and either of the troponins may optimize early identification of high-risk patients so they may be targeted for a higher level of care and consideration of more aggressive therapies.

Evaluation of alternative spectral feature extraction methods of textural images for multivariate modelling

SUMMARY Fast and automatic strategies for extraction of characteristic feature spectra from digital images are investigated. We present a study based on images from confocal laser scanning microscopy (CLSM) of mayonnaise. Based on principal component regression (PCR), six different methods are compared with respect to prediction of external measurements describing the sensory texture of samples. The methods considered are: 1, the magnitude spectrum of the Fourier transform; 2, the autocorrelation spectrum; 3, the autocovariance spectrum; 4, the absolute difference spectrum;5, the singular value spectrum;6, the angle measure technique.A technique based on crossvalidated predictions combined with a two-way ANOVA is suggested to decide eventual differences in prediction ability. © 1998 John Wiley & Sons, Ltd.

Evaluation of alternative spectral feature extraction methods of textural images for multivariate modelling

Fast and automatic strategies for extraction of characteristic feature spectra from digital images are investigated. We present a study based on images from confocal laser scanning microscopy (CLSM) of mayonnaise. Based on principal component regression (PCR), six different methods are compared with respect to prediction of external measurements describing the sensory texture of samples. The methods considered are: 1, the magnitude spectrum of the Fourier transform; 2, the autocorrelation spectrum; 3, the autocovariance spectrum; 4, the absolute difference spectrum; 5, the singular value spectrum; 6, the angle measure technique. A technique based on cross-validated predictions combined with a two-way ANOVA is suggested to decide eventual differences in prediction ability. © 1998 John Wiley & Sons, Ltd.