基于IRT推断识字量的方法研究

On the IRT-based Literacy Quantity Inference Method

为了实现基于IRT推断识字量,本研究模拟了一批识字量可知的虚拟被试,并建立了识字能力Logit分数与识字量的对应关系。研究发现虚拟被试的识字能力能够与其识字量建立起具有高解释率的正态卵形曲线对应关系。此外,基于虚拟被试和真实被试的作答矩阵,分别估计的真实被试的识字能力具有高相关,这一定程度上说明了模拟虚拟被试方式的合理性。最后,本研究比较了基于IRT和基于CTT的识字量推断方式的估计精度,结果表明,基于IRT的识字量推断方式能够有效地提高识字量的估计精度。

In order to realize IRT-based literacy quantity inference,this paper simulated a number of virtual subjects with known literacy and established the corresponding relationship between Logit score and literacy. It is found that the literacy parameters of virtual subjects can establish a normal oval curve with high interpretation rate. In addition,the measurement ruler constructed by the response matrix of virtual subjects and real subjects respectively estimated that the literacy ability of real subjects was highly correlated,which explained the rationality of simulating virtual subjects to some extent. Finally,based on the first two results,this study compared the estimation accuracy of the IRT-based and CTT-based literacy inference method. The results showed that the IRT-based literacy quantity inference method can improve the estimation accuracy of literacy.