[{"data":1,"prerenderedAt":447},["ShallowReactive",2],{"content-query-6ZtnHrRoPS":3},{"_path":4,"_dir":5,"_draft":6,"_partial":6,"_locale":7,"title":8,"description":9,"layout":10,"body":11,"_type":441,"_id":442,"_source":443,"_file":444,"_extension":445,"sitemap":446},"/documentation/algorithms-and-techniques/resampling-and-inference/bootstrapping","resampling-and-inference",false,"","Bootstrapping in PLS-SEM: Significance Testing in SmartPLS","Bootstrapping is a nonparametric resampling procedure for testing the significance of PLS-SEM results such as path coefficients, outer loadings, and HTMT. Learn how it works in SmartPLS.","algorithm-description",{"type":12,"children":13,"toc":424},"root",[14,23,29,36,41,46,52,54,175,182,202,219,225,231,236,242,247,253,258,264,269,275,280,286,337,343],{"type":15,"tag":16,"props":17,"children":19},"element","h1",{"id":18},"bootstrapping",[20],{"type":21,"value":22},"text","Bootstrapping",{"type":15,"tag":24,"props":25,"children":26},"p",{},[27],{"type":21,"value":28},"Bootstrapping is a nonparametric resampling procedure that allows researchers to test the statistical significance of PLS-SEM results such as path coefficients, outer weights, outer loadings, Cronbach's alpha, HTMT, and R² values. Because PLS-SEM does not assume that data are normally distributed, the parametric significance tests used in regression analysis cannot be applied to test whether these estimates are significant. Bootstrapping addresses this by repeatedly re-estimating the model on resampled data to derive standard errors, t-values, and confidence intervals for significance testing. Becker et al. (2023) and Hair et al. (2022) explain bootstrapping in PLS-SEM in more detail.",{"type":15,"tag":30,"props":31,"children":33},"h2",{"id":32},"how-bootstrapping-works",[34],{"type":21,"value":35},"How Bootstrapping Works",{"type":15,"tag":24,"props":37,"children":38},{},[39],{"type":21,"value":40},"In bootstrapping, subsamples are created from randomly drawn observations of the original data set (with replacement), following the nonparametric bootstrap procedure (Efron & Tibshirani, 1993; Davison & Hinkley, 1997). Each subsample is used to estimate the PLS path model, and this process is repeated until a large number of random subsamples has been created, typically about 10,000.",{"type":15,"tag":24,"props":42,"children":43},{},[44],{"type":21,"value":45},"The parameter estimates (e.g., outer weights, outer loadings, and path coefficients) obtained from the subsamples are used to derive 95% confidence intervals for significance testing: original PLS-SEM results are significant when they fall outside the confidence interval. In addition, bootstrapping provides the standard errors for the estimates, which allow t-values to be calculated to assess the significance of each estimate.",{"type":15,"tag":30,"props":47,"children":49},{"id":48},"bootstrapping-settings-in-smartpls",[50],{"type":21,"value":51},"Bootstrapping Settings in SmartPLS",{"type":21,"value":53},"\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n",{"type":15,"tag":55,"props":56,"children":57},"table",{},[58,82],{"type":15,"tag":59,"props":60,"children":61},"thead",{},[62],{"type":15,"tag":63,"props":64,"children":65},"tr",{},[66,72,77],{"type":15,"tag":67,"props":68,"children":69},"th",{},[70],{"type":21,"value":71},"Setting",{"type":15,"tag":67,"props":73,"children":74},{},[75],{"type":21,"value":76},"Default",{"type":15,"tag":67,"props":78,"children":79},{},[80],{"type":21,"value":81},"Description",{"type":15,"tag":83,"props":84,"children":85},"tbody",{},[86,105,123,141,158],{"type":15,"tag":63,"props":87,"children":88},{},[89,95,100],{"type":15,"tag":90,"props":91,"children":92},"td",{},[93],{"type":21,"value":94},"Subsamples",{"type":15,"tag":90,"props":96,"children":97},{},[98],{"type":21,"value":99},"10,000",{"type":15,"tag":90,"props":101,"children":102},{},[103],{"type":21,"value":104},"Number of bootstrap subsamples drawn with replacement from the original data set. Each subsample has the same number of observations as the original sample (or the reduced sample when case-by-case deletion is used for missing values). Larger numbers of subsamples improve result stability but increase computation time; 1,000 subsamples may suffice for an initial assessment, while 10,000 is recommended for final results.",{"type":15,"tag":63,"props":106,"children":107},{},[108,113,118],{"type":15,"tag":90,"props":109,"children":110},{},[111],{"type":21,"value":112},"Do parallel processing",{"type":15,"tag":90,"props":114,"children":115},{},[116],{"type":21,"value":117},"—",{"type":15,"tag":90,"props":119,"children":120},{},[121],{"type":21,"value":122},"Runs the bootstrapping algorithm on multiple processors, if available, since each subsample can be calculated independently. Reduces computation time.",{"type":15,"tag":63,"props":124,"children":125},{},[126,131,136],{"type":15,"tag":90,"props":127,"children":128},{},[129],{"type":21,"value":130},"Confidence interval method",{"type":15,"tag":90,"props":132,"children":133},{},[134],{"type":21,"value":135},"Percentile bootstrap",{"type":15,"tag":90,"props":137,"children":138},{},[139],{"type":21,"value":140},"Method used to estimate nonparametric confidence intervals: percentile bootstrap (default), studentized bootstrap, or bias-corrected and accelerated (BCa) bootstrap. Percentile bootstrapping is recommended by default; BCa bootstrapping is an alternative if the bootstrap distribution is a concern (Hair et al., 2022).",{"type":15,"tag":63,"props":142,"children":143},{},[144,149,153],{"type":15,"tag":90,"props":145,"children":146},{},[147],{"type":21,"value":148},"Test type",{"type":15,"tag":90,"props":150,"children":151},{},[152],{"type":21,"value":117},{"type":15,"tag":90,"props":154,"children":155},{},[156],{"type":21,"value":157},"Specifies whether a one-sided or two-sided significance test is conducted.",{"type":15,"tag":63,"props":159,"children":160},{},[161,166,170],{"type":15,"tag":90,"props":162,"children":163},{},[164],{"type":21,"value":165},"Significance level",{"type":15,"tag":90,"props":167,"children":168},{},[169],{"type":21,"value":117},{"type":15,"tag":90,"props":171,"children":172},{},[173],{"type":21,"value":174},"Specifies the significance level of the test statistic.",{"type":15,"tag":176,"props":177,"children":179},"h3",{"id":178},"random-number-generator",[180],{"type":21,"value":181},"Random Number Generator",{"type":15,"tag":24,"props":183,"children":184},{},[185,187,193,195,200],{"type":21,"value":186},"Because bootstrapping randomly generates subsamples from the original data set, it requires a seed value for the random number generator. You can choose between a ",{"type":15,"tag":188,"props":189,"children":190},"em",{},[191],{"type":21,"value":192},"random seed",{"type":21,"value":194}," and a ",{"type":15,"tag":188,"props":196,"children":197},{},[198],{"type":21,"value":199},"fixed seed",{"type":21,"value":201},".",{"type":15,"tag":24,"props":203,"children":204},{},[205,207,211,213,217],{"type":21,"value":206},"The ",{"type":15,"tag":188,"props":208,"children":209},{},[210],{"type":21,"value":192},{"type":21,"value":212}," produces different random numbers, and therefore different results, every time the algorithm is executed (this was the default and only option in SmartPLS 3). The ",{"type":15,"tag":188,"props":214,"children":215},{},[216],{"type":21,"value":199},{"type":21,"value":218}," uses a pre-specified seed value that stays the same across executions, so it produces the same results whenever the same number of subsamples is drawn, addressing concerns about the replicability of research findings.",{"type":15,"tag":30,"props":220,"children":222},{"id":221},"frequently-asked-questions",[223],{"type":21,"value":224},"Frequently Asked Questions",{"type":15,"tag":176,"props":226,"children":228},{"id":227},"why-does-pls-sem-need-bootstrapping-instead-of-parametric-significance-tests",[229],{"type":21,"value":230},"Why does PLS-SEM need bootstrapping instead of parametric significance tests?",{"type":15,"tag":24,"props":232,"children":233},{},[234],{"type":21,"value":235},"PLS-SEM does not assume that the data are normally distributed, so the parametric significance tests used in regression analysis cannot be applied to coefficients such as outer weights, outer loadings, and path coefficients. Bootstrapping is a nonparametric alternative that derives standard errors, t-values, and confidence intervals directly from resampled data.",{"type":15,"tag":176,"props":237,"children":239},{"id":238},"how-many-bootstrap-subsamples-should-i-use",[240],{"type":21,"value":241},"How many bootstrap subsamples should I use?",{"type":15,"tag":24,"props":243,"children":244},{},[245],{"type":21,"value":246},"For an initial assessment, a smaller number such as 1,000 subsamples is often sufficient and faster to compute. For final results preparation, use a large number of subsamples, typically about 10,000, to ensure stable results.",{"type":15,"tag":176,"props":248,"children":250},{"id":249},"which-confidence-interval-method-should-i-choose",[251],{"type":21,"value":252},"Which confidence interval method should I choose?",{"type":15,"tag":24,"props":254,"children":255},{},[256],{"type":21,"value":257},"Percentile bootstrapping is the recommended default. If you have concerns about a non-normal bootstrap distribution, bias-corrected and accelerated (BCa) bootstrapping is a suitable alternative. Studentized bootstrapping is also available.",{"type":15,"tag":176,"props":259,"children":261},{"id":260},"what-is-the-difference-between-a-random-seed-and-a-fixed-seed",[262],{"type":21,"value":263},"What is the difference between a random seed and a fixed seed?",{"type":15,"tag":24,"props":265,"children":266},{},[267],{"type":21,"value":268},"A random seed produces different results every time bootstrapping is run, which was the only option in SmartPLS 3. A fixed seed uses the same pre-specified seed value across executions, producing identical results whenever the same number of subsamples is drawn, which supports the replicability of findings.",{"type":15,"tag":176,"props":270,"children":272},{"id":271},"does-using-more-bootstrap-subsamples-change-my-results",[273],{"type":21,"value":274},"Does using more bootstrap subsamples change my results?",{"type":15,"tag":24,"props":276,"children":277},{},[278],{"type":21,"value":279},"More subsamples improve the stability of the estimated standard errors and confidence intervals, but they also increase computation time. Using too few subsamples can make results less reliable, which is why a large number (e.g., 10,000) is recommended for the final analysis.",{"type":15,"tag":30,"props":281,"children":283},{"id":282},"related-smartpls-methods",[284],{"type":21,"value":285},"Related SmartPLS Methods",{"type":15,"tag":287,"props":288,"children":289},"ul",{},[290,301,310,319,328],{"type":15,"tag":291,"props":292,"children":293},"li",{},[294],{"type":15,"tag":295,"props":296,"children":298},"a",{"href":297},"/documentation/algorithms-and-techniques/resampling-and-inference/consistent-bootstrapping/",[299],{"type":21,"value":300},"Consistent bootstrapping",{"type":15,"tag":291,"props":302,"children":303},{},[304],{"type":15,"tag":295,"props":305,"children":307},{"href":306},"/documentation/algorithms-and-techniques/regression-bootstrapping/",[308],{"type":21,"value":309},"Regression bootstrapping",{"type":15,"tag":291,"props":311,"children":312},{},[313],{"type":15,"tag":295,"props":314,"children":316},{"href":315},"/documentation/algorithms-and-techniques/heterogeneity-and-multigroup/multigroup-analysis/",[317],{"type":21,"value":318},"Multigroup analysis (bootstrap MGA)",{"type":15,"tag":291,"props":320,"children":321},{},[322],{"type":15,"tag":295,"props":323,"children":325},{"href":324},"/documentation/functionalities/thresholds/",[326],{"type":21,"value":327},"Result color thresholds",{"type":15,"tag":291,"props":329,"children":330},{},[331],{"type":15,"tag":295,"props":332,"children":334},{"href":333},"/documentation/algorithms-and-techniques/core-algorithm/pls/",[335],{"type":21,"value":336},"PLS-SEM algorithm",{"type":15,"tag":30,"props":338,"children":340},{"id":339},"references",[341],{"type":21,"value":342},"References",{"type":15,"tag":287,"props":344,"children":345},{},[346,375,391,403,415],{"type":15,"tag":291,"props":347,"children":348},{},[349,351,359,361,366,368,373],{"type":21,"value":350},"Becker, J.-M., Cheah, J. H., Gholamzade, R., Ringle, C. M., & Sarstedt, M. (2023). ",{"type":15,"tag":295,"props":352,"children":356},{"href":353,"rel":354},"https://www.emerald.com/insight/content/doi/10.1108/IJCHM-04-2022-0474/full/html",[355],"nofollow",[357],{"type":21,"value":358},"PLS-SEM's most wanted guidance.",{"type":21,"value":360}," ",{"type":15,"tag":188,"props":362,"children":363},{},[364],{"type":21,"value":365},"International Journal of Contemporary Hospitality Management",{"type":21,"value":367},", ",{"type":15,"tag":188,"props":369,"children":370},{},[371],{"type":21,"value":372},"35",{"type":21,"value":374},"(1), 321–346.",{"type":15,"tag":291,"props":376,"children":377},{},[378,380,389],{"type":21,"value":379},"Hair, J. F., Hult, G. T. M., Ringle, C. M., & Sarstedt, M. (2027). ",{"type":15,"tag":188,"props":381,"children":382},{},[383],{"type":15,"tag":295,"props":384,"children":386},{"href":385},"/documentation/must-reads/pls-sem-book",[387],{"type":21,"value":388},"A primer on partial least squares structural equation modeling (PLS-SEM)",{"type":21,"value":390}," (4th ed.). Sage.",{"type":15,"tag":291,"props":392,"children":393},{},[394,396,401],{"type":21,"value":395},"Davison, A. C., & Hinkley, D. V. (1997). ",{"type":15,"tag":188,"props":397,"children":398},{},[399],{"type":21,"value":400},"Bootstrap methods and their application",{"type":21,"value":402},". Cambridge University Press.",{"type":15,"tag":291,"props":404,"children":405},{},[406,408,413],{"type":21,"value":407},"Efron, B., & Tibshirani, R. J. (1993). ",{"type":15,"tag":188,"props":409,"children":410},{},[411],{"type":21,"value":412},"An introduction to the bootstrap",{"type":21,"value":414},". Chapman & Hall.",{"type":15,"tag":291,"props":416,"children":417},{},[418],{"type":15,"tag":295,"props":419,"children":421},{"href":420},"/documentation",[422],{"type":21,"value":423},"More literature ...",{"title":7,"searchDepth":425,"depth":425,"links":426},2,[427,428,432,439,440],{"id":32,"depth":425,"text":35},{"id":48,"depth":425,"text":51,"children":429},[430],{"id":178,"depth":431,"text":181},3,{"id":221,"depth":425,"text":224,"children":433},[434,435,436,437,438],{"id":227,"depth":431,"text":230},{"id":238,"depth":431,"text":241},{"id":249,"depth":431,"text":252},{"id":260,"depth":431,"text":263},{"id":271,"depth":431,"text":274},{"id":282,"depth":425,"text":285},{"id":339,"depth":425,"text":342},"markdown","content:documentation:algorithms-and-techniques:resampling-and-inference:bootstrapping:index.md","content","documentation/algorithms-and-techniques/resampling-and-inference/bootstrapping/index.md","md",{"loc":4},1784805332054]