[{"data":1,"prerenderedAt":344},["ShallowReactive",2],{"content-query-REckbw1fme":3},{"_path":4,"_dir":5,"_draft":6,"_partial":6,"_locale":7,"title":8,"description":9,"layout":10,"body":11,"_type":338,"_id":339,"_source":340,"_file":341,"_extension":342,"sitemap":343},"/documentation/algorithms-and-techniques/resampling-and-inference/blindfolding","resampling-and-inference",false,"","Blindfolding and Stone-Geisser's Q² in PLS-SEM","Blindfolding produces Stone-Geisser's Q² to assess the predictive relevance of a PLS-SEM model. Learn how the procedure works and why SmartPLS 4 recommends PLSpredict and CVPAT instead.","algorithm-description",{"type":12,"children":13,"toc":321},"root",[14,23,29,36,43,48,53,60,65,70,75,81,83,134,139,145,151,156,162,167,173,178,184,189,195,200,206,248,254],{"type":15,"tag":16,"props":17,"children":19},"element","h1",{"id":18},"blindfolding",[20],{"type":21,"value":22},"text","Blindfolding",{"type":15,"tag":24,"props":25,"children":26},"p",{},[27],{"type":21,"value":28},"Blindfolding is a sample re-use technique that produces Stone-Geisser's Q² value (Stone, 1974; Geisser, 1974), a criterion for judging the cross-validated predictive relevance of a PLS path model's endogenous constructs. Researchers traditionally examined Q² alongside R² to assess how well a PLS-SEM model explains and predicts its indicators. Because blindfolding only provides an in-sample assessment and does not test out-of-sample predictive power, SmartPLS 4 has discontinued the algorithm in favor of PLSpredict and the cross-validated predictive ability test (CVPAT), described below.",{"type":15,"tag":30,"props":31,"children":33},"div",{"style":32},"background-color:red;color:white;padding:10px;",[34],{"type":21,"value":35},"We have discontinued support for blindfolding in SmartPLS 4 and removed the algorithm. The blindfolding method does not provide an out-of-sample assessment of predictive power. However, the PLSpredict procedure and the cross-validated predictive ability test (CVPAT) provide the results required for an out-of-sample predictive power assessment (for further explanations, see Hair et al., 2022). [PLSpredict](/documentation/algorithms-and-techniques/prediction-and-segmentation/predict) and the [CVPAT](/documentation/algorithms-and-techniques/resampling-and-inference/cvpat) have been implemented in SmartPLS and we recommend using these methods instead of blindfolding.",{"type":15,"tag":37,"props":38,"children":40},"h2",{"id":39},"stone-geissers-q-and-predictive-relevance",[41],{"type":21,"value":42},"Stone-Geisser's Q² and Predictive Relevance",{"type":15,"tag":24,"props":44,"children":45},{},[46],{"type":21,"value":47},"Besides evaluating the magnitude of R² as a criterion of predictive accuracy, researchers may also want to examine Stone-Geisser's Q² value (Stone, 1974; Geisser, 1974) as a criterion of predictive relevance. The Q² value of latent variables in a PLS path model is obtained through the blindfolding procedure.",{"type":15,"tag":24,"props":49,"children":50},{},[51],{"type":21,"value":52},"When PLS-SEM exhibits predictive relevance, it accurately predicts the data points of its indicators. A Q² value larger than zero for a given endogenous latent variable indicates that the PLS path model has predictive relevance for this construct. For detailed explanations of the blindfolding procedure, see Hair et al. (2017).",{"type":15,"tag":54,"props":55,"children":57},"h3",{"id":56},"how-the-blindfolding-procedure-works",[58],{"type":21,"value":59},"How the Blindfolding Procedure Works",{"type":15,"tag":24,"props":61,"children":62},{},[63],{"type":21,"value":64},"Blindfolding is a sample re-use technique that systematically deletes data points and produces a prognosis of their original values. The procedure requires an omission distance D, for which the literature recommends a value between 5 and 12 (e.g., Hair et al., 2017).",{"type":15,"tag":24,"props":66,"children":67},{},[68],{"type":21,"value":69},"In the first blindfolding round, the procedure starts with the first data point and omits every D-th data point of a latent variable's indicators, then estimates the PLS path model using the remaining data points. The omitted data are treated as missing values (e.g., through mean value replacement or pairwise deletion). The PLS-SEM results are then used to predict the omitted data points, and the difference between the omitted and predicted values is the prediction error. The sum of squared prediction errors is used to calculate the Q² value.",{"type":15,"tag":24,"props":71,"children":72},{},[73],{"type":21,"value":74},"Blindfolding is an iterative process: in the second round, the algorithm starts with the second data point and again omits every D-th data point, and so on. Because the procedure must omit and predict every data point of the indicators used in the measurement model of the selected latent variable, the number of blindfolding rounds always equals the omission distance D. For example, an omission distance of seven (D=7) results in seven blindfolding rounds, with every seventh data point eliminated in each round. After D rounds, every data point has been omitted and predicted once.",{"type":15,"tag":37,"props":76,"children":78},{"id":77},"blindfolding-settings-in-smartpls",[79],{"type":21,"value":80},"Blindfolding Settings in SmartPLS",{"type":21,"value":82},"\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n",{"type":15,"tag":84,"props":85,"children":86},"table",{},[87,111],{"type":15,"tag":88,"props":89,"children":90},"thead",{},[91],{"type":15,"tag":92,"props":93,"children":94},"tr",{},[95,101,106],{"type":15,"tag":96,"props":97,"children":98},"th",{},[99],{"type":21,"value":100},"Setting",{"type":15,"tag":96,"props":102,"children":103},{},[104],{"type":21,"value":105},"Default",{"type":15,"tag":96,"props":107,"children":108},{},[109],{"type":21,"value":110},"Description",{"type":15,"tag":112,"props":113,"children":114},"tbody",{},[115],{"type":15,"tag":92,"props":116,"children":117},{},[118,124,129],{"type":15,"tag":119,"props":120,"children":121},"td",{},[122],{"type":21,"value":123},"Omission distance (D)",{"type":15,"tag":119,"props":125,"children":126},{},[127],{"type":21,"value":128},"7",{"type":15,"tag":119,"props":130,"children":131},{},[132],{"type":21,"value":133},"Determines the systematic pattern of data point elimination and prediction. Suggested values are between 5 and 12 (Hair et al., 2017). The number of blindfolding rounds always equals D.",{"type":15,"tag":24,"props":135,"children":136},{},[137],{"type":21,"value":138},"The omission distance D must be chosen so that the number of observations in the data set divided by D is not an integer. If the division results in an integer, the procedure would delete entire rows of the data set, so fewer observations would be used per blindfolding round than are available in the original data set. Since the goal of blindfolding is to use all observations for prediction rather than discard entire observations, the number of observations divided by D must not be an integer.",{"type":15,"tag":37,"props":140,"children":142},{"id":141},"frequently-asked-questions",[143],{"type":21,"value":144},"Frequently Asked Questions",{"type":15,"tag":54,"props":146,"children":148},{"id":147},"why-has-smartpls-discontinued-blindfolding",[149],{"type":21,"value":150},"Why has SmartPLS discontinued blindfolding?",{"type":15,"tag":24,"props":152,"children":153},{},[154],{"type":21,"value":155},"Blindfolding only provides an in-sample assessment of predictive relevance and does not test a model's out-of-sample predictive power. SmartPLS 4 removed the algorithm and instead recommends PLSpredict and the cross-validated predictive ability test (CVPAT), which do provide out-of-sample predictive power assessments.",{"type":15,"tag":54,"props":157,"children":159},{"id":158},"what-does-stone-geissers-q-value-measure",[160],{"type":21,"value":161},"What does Stone-Geisser's Q² value measure?",{"type":15,"tag":24,"props":163,"children":164},{},[165],{"type":21,"value":166},"Q² is a criterion for the cross-validated predictive relevance of a PLS path model's endogenous constructs. A Q² value larger than zero for a given endogenous latent variable indicates that the model has predictive relevance for that construct.",{"type":15,"tag":54,"props":168,"children":170},{"id":169},"what-is-the-omission-distance-in-blindfolding",[171],{"type":21,"value":172},"What is the omission distance in blindfolding?",{"type":15,"tag":24,"props":174,"children":175},{},[176],{"type":21,"value":177},"The omission distance (D) determines how many data points are systematically omitted and predicted in each blindfolding round. Recommended values are between 5 and 12, and the number of blindfolding rounds always equals D.",{"type":15,"tag":54,"props":179,"children":181},{"id":180},"why-cant-the-omission-distance-divide-the-number-of-observations-evenly",[182],{"type":21,"value":183},"Why can't the omission distance divide the number of observations evenly?",{"type":15,"tag":24,"props":185,"children":186},{},[187],{"type":21,"value":188},"If the number of observations divided by D is an integer, blindfolding would delete entire rows of the data set instead of individual data points, reducing the number of observations available per round. Choosing a D that does not evenly divide the sample size ensures all observations remain available for prediction.",{"type":15,"tag":54,"props":190,"children":192},{"id":191},"what-should-i-use-instead-of-blindfolding-in-smartpls-4",[193],{"type":21,"value":194},"What should I use instead of blindfolding in SmartPLS 4?",{"type":15,"tag":24,"props":196,"children":197},{},[198],{"type":21,"value":199},"SmartPLS recommends PLSpredict for out-of-sample predictive assessment and the cross-validated predictive ability test (CVPAT) for prediction-oriented model comparison, since both provide the out-of-sample results that blindfolding cannot.",{"type":15,"tag":37,"props":201,"children":203},{"id":202},"related-smartpls-methods",[204],{"type":21,"value":205},"Related SmartPLS Methods",{"type":15,"tag":207,"props":208,"children":209},"ul",{},[210,221,230,239],{"type":15,"tag":211,"props":212,"children":213},"li",{},[214],{"type":15,"tag":215,"props":216,"children":218},"a",{"href":217},"/documentation/algorithms-and-techniques/prediction-and-segmentation/predict/",[219],{"type":21,"value":220},"PLSpredict",{"type":15,"tag":211,"props":222,"children":223},{},[224],{"type":15,"tag":215,"props":225,"children":227},{"href":226},"/documentation/algorithms-and-techniques/resampling-and-inference/cvpat/",[228],{"type":21,"value":229},"Cross-validated predictive ability test (CVPAT)",{"type":15,"tag":211,"props":231,"children":232},{},[233],{"type":15,"tag":215,"props":234,"children":236},{"href":235},"/documentation/algorithms-and-techniques/resampling-and-inference/bootstrapping/",[237],{"type":21,"value":238},"Bootstrapping",{"type":15,"tag":211,"props":240,"children":241},{},[242],{"type":15,"tag":215,"props":243,"children":245},{"href":244},"/documentation/functionalities/thresholds/",[246],{"type":21,"value":247},"Result color thresholds",{"type":15,"tag":37,"props":249,"children":251},{"id":250},"references",[252],{"type":21,"value":253},"References",{"type":15,"tag":207,"props":255,"children":256},{},[257,277,294,312],{"type":15,"tag":211,"props":258,"children":259},{},[260,262,268,270,275],{"type":21,"value":261},"Geisser, S. (1974). A predictive approach to the random effects model. ",{"type":15,"tag":263,"props":264,"children":265},"em",{},[266],{"type":21,"value":267},"Biometrika",{"type":21,"value":269},", ",{"type":15,"tag":263,"props":271,"children":272},{},[273],{"type":21,"value":274},"61",{"type":21,"value":276},"(1), 101–107.",{"type":15,"tag":211,"props":278,"children":279},{},[280,282,292],{"type":21,"value":281},"Hair, J. F., Hult, G. T. M., Ringle, C. M., & Sarstedt, M. (2027). ",{"type":15,"tag":215,"props":283,"children":285},{"href":284},"/documentation/must-reads/pls-sem-book",[286],{"type":15,"tag":287,"props":288,"children":289},"strong",{},[290],{"type":21,"value":291},"A Primer on Partial Least Squares Structural Equation Modeling (PLS-SEM)",{"type":21,"value":293}," (4th ed.). Sage.",{"type":15,"tag":211,"props":295,"children":296},{},[297,299,304,305,310],{"type":21,"value":298},"Stone, M. (1974). Cross-validatory choice and assessment of statistical predictions. ",{"type":15,"tag":263,"props":300,"children":301},{},[302],{"type":21,"value":303},"Journal of the Royal Statistical Society",{"type":21,"value":269},{"type":15,"tag":263,"props":306,"children":307},{},[308],{"type":21,"value":309},"36",{"type":21,"value":311},"(2), 111–147.",{"type":15,"tag":211,"props":313,"children":314},{},[315],{"type":15,"tag":215,"props":316,"children":318},{"href":317},"/documentation",[319],{"type":21,"value":320},"More literature ...",{"title":7,"searchDepth":322,"depth":322,"links":323},2,[324,328,329,336,337],{"id":39,"depth":322,"text":42,"children":325},[326],{"id":56,"depth":327,"text":59},3,{"id":77,"depth":322,"text":80},{"id":141,"depth":322,"text":144,"children":330},[331,332,333,334,335],{"id":147,"depth":327,"text":150},{"id":158,"depth":327,"text":161},{"id":169,"depth":327,"text":172},{"id":180,"depth":327,"text":183},{"id":191,"depth":327,"text":194},{"id":202,"depth":322,"text":205},{"id":250,"depth":322,"text":253},"markdown","content:documentation:algorithms-and-techniques:resampling-and-inference:blindfolding:index.md","content","documentation/algorithms-and-techniques/resampling-and-inference/blindfolding/index.md","md",{"loc":4},1784805332767]