How to Quantify Relative Importance of Multiple Predictors/Effect Sizes?

是 否 +2 论坛币

k人 参与回答

经管之家送您一份

应届毕业生专属福利!

求职就业群

赵安豆老师微信：zhaoandou666

经管之家联合CDA

送您一个全额奖学金名额~ !

立即领取

感谢您参与论坛问题回答

经管之家送您两个论坛币！

+2 论坛币

My quesion is  to find out which fixed factors (and cross-level interactions) had the strongest effects. I know as per Snijders and Bosker there are methods by which to extract effect sizes for these nested structures (e.g., using the variance components for  PRE statistic), but many examples I've seen are for single- predictor models.
Where I am stuck is how to rank-order magnitude of effect for individual variables when you have multiple predictors and/or cross-level interactions (i.e., time varying x time invariant interaction).  I'm assuming that similar to the problems encountered when rank ordering strength of explanatory variables in linear or logistic regression models,  one needs to factor in the extent of collinearity (which dominance analysis and relative weights does).

Somebody suggested standardizing all of the explanatory variables and then rank ordering (or squaring) the fixed standardized coefficients, but this does not address the extent of collinearity.  I was thinking of possibly freeing up the stochastic parameters for each of the fixed predictors and using those as a PRE statistics, but when I have done so for multiple predictors the model tends to implode (took over 1000 iterations in HLM7.0!).

So I was curious how to address the relative importance of multiple predictors/effect sizes (and partitioning of variance) for a random coefficient model,  and whether it be of a cross-sectional or longitudinal design?

Thank you very much.

扫码加我 拉你入群

请注明：姓名-公司-职位

以便审核进群资格，未注明则拒绝

分享0 收藏0 回帖

关键词：effect size importance Predictors predictor Multiple individual multiple relative examples address