潜变量、显变量和贝叶斯中介效应模型比较

徐洪吕; 苏莹珍; 陶剑; 郭纪昌; 陶芳标

doi:10.16462/j.cnki.zhjbkz.2023.08.013

潜变量、显变量和贝叶斯中介效应模型比较

doi: 10.16462/j.cnki.zhjbkz.2023.08.013

1.
昆明学院医学院临床医学系，昆明 650214
2.
兴义民族师范学院教育科学学院，兴义 562400
3.
安徽医科大学公共卫生学院儿少卫生与妇幼保健学系，合肥 230032

基金项目:

国家自然科学基金 82160622

云南省地方本科高校基础研究联合专项资金项目 202101BA070001-117

昆明学院引进人才项目 YJL2103

详细信息

通讯作者:
陶芳标，E-mail: fbtao@ahmu.edu.cn

中图分类号: R563.1; R181
计量
- 文章访问数: 452
- HTML全文浏览量: 234
- PDF下载量: 43
- 被引次数: 0
出版历程
- 收稿日期: 2022-03-11
- 修回日期: 2022-09-20
- 网络出版日期: 2023-09-02
- 刊出日期: 2023-08-10

A comparative analysis of comparison of latent variables, manifest variables and Bayesian mediation effect models

1.
Department of Clinical Medicine, School of Medicine, Kunming University, Kunming 650214, China
2.
School of Educational Science, Minzu Normal University of Xingyi, Xingyi 562400, China
3.
Department of Maternal, Child and Adolescent Health Care, School of Public Health, Anhui Medical University, Hefei 230032, China

Funds:

National Natural Science Foundation of China 82160622

The Special Basic Cooperative Research Programs of Yunnan Provincial Undergraduate Universities Association 202101BA070001-117

Introduced Talents Scientific Research Project of Kunming University YJL2103

More Information

Corresponding author: TAO Fangbiao, E-mail: fbtao@ahmu.edu.cn

摘要

摘要: 目的比较潜变量、显变量和贝叶斯中介效应模型的中介效应及模型拟合优劣。方法数据来源于一项关于大学生行为与健康的专项调查。6种饮料消费评分为自变量，匹兹堡睡眠质量指数量表 7个维度评分为中介变量，9项患者健康问卷评估抑郁症状评分为因变量。分别使用潜变量、显变量和贝叶斯中介效应模型分析睡眠质量在饮料消费与抑郁症状关联中的中介效应。结果 3种模型的中介效应值分别为0.12、0.06和0.06，中介效应分别占总效应的71%、43%和43%，中介效应与直接效应之比分别为2.49、0.76和0.76。显变量中介效应模型和贝叶斯中介效应模型的中介效应几乎一致，而潜变量中介效应模型的中介效应估计值、中介效应与总效应之比和中介效应与直接效应之比分别是前2种模型的2.00倍、1.65倍和3.27倍。结论 3种模型均显示存在中介效应，潜变量中介效应模型估计的中介效应值更高，贝叶斯中介效应模型拟合评价指标更丰富。
- 中介效应 /
- 结构方程模型 /
- 显变量 /
- 潜变量 /
- 贝叶斯
Abstract: Objective The study aimed to compare the mediating effects and model fitting of the latent variable, the manifest variable and the Bayesian mediating effect model. Methods Data from a previous specialized survey on college students′ behavior and health were utilized. Six beverage consumption scores served as independent variables, seven dimensions of the Pittsburgh Sleep Quality Index Scale were mediating variables, and nine patient health questionnaire assessment depressive symptom scores were dependent variables. The mediating effects of sleep quality on the association between beverage consumption and depressive symptoms were analyzed using latent variable, manifest variable, and Bayesian mediation effect models. Results The mediating effect values of the above three models were 0.12, 0.06, and 0.06, respectively. The mediating effect accounted for 71%, 43%, and 43% of the total effect, and the ratio of the mediating effect to the direct effect were 2.49, 0.76, and 0.76, respectively. The mediating effects of the manifest variable mediation effect model and the Bayesian mediation effect model were nearly identical. The estimated value of the mediating effect, the ratio of the mediating effect to the total effect, and the ratio of the mediating effect to the direct effect of the latent variable mediation effect model were 2.00 times, 1.65 times, and 3.27 times those of the other two models, respectively. Conclusions While the mediating effects of the three models were consistent, the latent variable mediation effect model estimated a higher mediating effect value, and the Bayesian mediation effect model provided a more comprehensive evaluation index for model fitting.
- Mediating effect /
- Structural equation model /
- Manifest variable /
- Latent variable /
- Bayesian

HTML全文

图 1 中介效应模型收敛检验

1. A: 后验参数分布图；2. B: 踪迹图；3. C: 自相关图。

Figure 1. Convergence test of the mediation effect model

1. A: posterior parameter distribution chart; 2. B: trace map; 3. C: autocorrelation diagram.

下载: 全尺寸图片幻灯片

图 2 贝叶斯中介效应模型拟合评价

1. A: 后验预测检验散点图；2. B: 后验检验直方图。

Figure 2. Evaluation of the fit of the Bayesian mediation effect model

1. A: posterior predictive test scatterplot; 2. B: post-test histogram.

下载: 全尺寸图片幻灯片

图 3 三种中介效应模型中介效应分析

1. A: 潜变量中介效应模型; 2. B: 显变量中介效应模型; 3. C: 贝叶斯中介效应模型; 4. a: 总效应; 5. b: 中介效应与总效应之比; 6. c: 中介效应与直接效应之比。

Figure 3. Analysis of the mediation effect of three mediation effect models

1. A: latent variable mediation effect model; 2. B: manifest variable mediation effect model; 3. C: Bayesian mediation effect model; 4. a: total effect; 5. b: the ratio of the mediating effect to the total effect; 6. c: the ratio of mediating effects to direct effects.

下载: 全尺寸图片幻灯片

图 4 贝叶斯中介效应模型拟合评价

1. A: 潜变量中介效应模型; 2. B: 显变量中介效应模型; 3. C: 贝叶斯中介效应模型; 4. x: 饮料消费; 5. m: 睡眠质量; 6. y: 抑郁症状。

Figure 4. Evaluation of the fit of the Bayesian mediation effect model

1. A: latent variable mediation effect model; 2. B: manifest variable mediation effect model; 3. C: Bayesian mediation effect model; 4. x: beverage consumption; 5. m: sleep quality; 6. y: depressive symptoms.

下载: 全尺寸图片幻灯片

表 1 变量赋值及基本情况

Table 1. Variable assignment and basic information

模型Model	AIC	BIC	样本量调整BIC Sample-size adjusted BIC	模型拟合χ²检验χ²值 Chi-Square test of model fit Chi-Square value	模型拟合χ²检验P值 Chi-Square test of model fit P value	RMSEA
潜变量中介效应模型 Latent variable mediating effect model	223 988.90	224 458.95	224 226.98	3 829.27	0.00	0.06
显变量中介效应模型 Manifest variable mediating effect model	49 402.42	49 447.49	49 425.25	0.00	0.00	0.00
贝叶斯中介效应模型 Bayesian mediating effect model	―	49 447.50	―	―	―	―

模型Model	CFI	TLI	SRMR	贝叶斯后验预测检验χ²值 Bayesian posterior predictive phecking using Chi-Square	DIC	估计参数数量(pD) Estimated Number of Parameters (pD)
潜变量中介效应模型 Latent variable mediating effect model	0.92	0.90	0.05	―	―	―
显变量中介效应模型 Manifest variable mediating effect model	1.00	1.00	0.00	―	―	―
贝叶斯中介效应模型 Bayesian mediating effect model	―	―	―	0.51	49 402.58	7.08
注：1. “―”表示模型未提供数据。 2. AIC, 赤池信息准则; BIC, 贝叶斯信息准则; RMSEA, 近似误差均方根; CFI, 比较拟合指数; TLI, Tucker-Lewis指数; SRMR, 标准化残差均方和平方根; DIC, 偏差信息准则。 Note: 1. "―" indicates that the model did not provide data. 2. AIC, Akaike information criterion; BIC, Bayesian information criterion; RMSEA, root-mean-square error of approximation; CFI, comparative fit index; TLI, Tucker-Lewis index; SRMR, standardized root mean square residual; DIC, deviance information criterion.

下载: 导出CSV

表 2 3种模型中介效应估计值

Table 2. Estimates of the mediation effect of the three models

模型Model	效应 Effect	路径Path	95% CI	效应值 Effect value	SE/SD ^①	t值 value	P值 value
潜变量中介效应模型 Latent variable mediating effect model	直接效应 Direct effect	饮料消费→抑郁症状 Beverage consumption→Depressive symptoms	0.02~0.08	0.05	0.02	2.88	0.004
		睡眠质量→抑郁症状 Sleep quality→Depressive symptoms	0.73~0.78	0.76	0.02	50.30	< 0.001
		饮料消费→睡眠质量 Beverage consumption→Sleep quality	0.11~0.21	0.16	0.03	6.01	< 0.001
	中介效应 Mediating effect	饮料消费→睡眠质量→抑郁症状 Beverage consumption→Sleep quality→Depressive symptoms	0.08~0.16	0.12	0.02	5.91	< 0.001
显变量中介效应模型 Manifest variable mediating effect model	直接效应 Direct effect	饮料消费→抑郁症状 Beverage consumption→Depressive symptoms	0.05~0.11	0.08	0.02	5.32	< 0.001
		睡眠质量→抑郁症状 Sleep quality→Depressive symptoms	0.54~0.59	0.56	0.01	44.46	< 0.001
		饮料消费→睡眠质量 Beverage consumption→Sleep quality	0.08~0.15	0.11	0.02	6.27	< 0.001
	中介效应 Mediating effect	饮料消费→睡眠质量→抑郁症状 Beverage consumption→Sleep quality→Depressive symptoms	0.04~0.08	0.06	0.01	6.24	< 0.001
贝叶斯中介效应模型 Bayesian mediating effect model	直接效应 Direct effect	饮料消费→抑郁症状 Beverage consumption→Depressive symptoms	0.06~0.11	0.01	—	0.08	< 0.001
		睡眠质量→抑郁症状 Sleep quality→Depressive symptoms	0.54~0.58	0.56	0.01	—	< 0.001
		饮料消费→睡眠质量 Beverage consumption→Sleep quality	0.08~0.14	0.11	0.01	—	< 0.001
	中介效应 Mediating effect	饮料消费→睡眠质量→抑郁症状 Beverage consumption→Sleep quality→Depressive symptoms	0.05~0.08	0.06	0.01	—	< 0.001
注：“—”表示模型未提供数据。 ①潜变量中介效应模型和显变量中介效应模型提供SE，贝叶斯中介效应模型提供SD。 Note: "—"indicates that the model did not provide data. ① The latent variable mediation effect model and the manifest variable mediation effect model provide SE, and the Bayesian mediation effect model provides SD.

下载: 导出CSV

参考文献(39)

[1]	Wang YB, Chen Z, Goldstein JM, et al. A Bayesian regularized mediation analysis with multiple exposures[J]. Stat Med, 2019, 38(5): 828-843. DOI: 10.1002/sim.8020.
[2]	温忠麟, 叶宝娟. 中介效应分析: 方法和模型发展[J]. 心理科学进展, 2014, 22(5): 731-745. DOI: 10.3724/SP.J.1042.2014.00731. Wen ZL, Ye BJ. Analyses of mediating effects: the development of methods and models[J]. Adv Psychol Sci, 2014, 22(5): 731-745. DOI: 10.3724/SP.J.1042.2014.00731.
[3]	温忠麟, 张雷, 侯杰泰, 等. 中介效应检验程序及其应用[J]. 心理学报, 2004, 36(5): 614-620. DOI: 10.3969/j.issn.1671-6981.2003.01.021. Wen ZL, Zhang L, Hou JT, et al. Testing and application of the mediating effects[J]. Acta Psychol Sin, 2004, 36(5): 614-620. DOI: 10.3969/j.issn.1671-6981.2003.01.021.
[4]	Henderson KA, Obeid N, Buchholz A, et al. Coping in adolescents: a mediator between stress and disordered eating[J]. Eat Behav, 2022, 47: 101626. DOI: 10.1016/j.eatbeh.2022.101626.
[5]	Xu H, Guo J, Wan Y, et al. Association between screen time, fast foods, sugar-sweetened beverages and depressive symptoms in Chinese adolescents[J]. Front Psychiatry, 2020, 11: 458. DOI: 10.3389/fpsyt.2020.00458.
[6]	Rijnhart JJM, Twisk JWR, Chinapaw MJM, et al. Comparison of methods for the analysis of relatively simple mediation models[J]. Contemp Clin Trials Commun, 2017, 7: 130-135. DOI: 10.1016/j.conctc.2017.06.005.
[7]	VanderWeele TJ, Tchetgen Tchetgen EJ. Mediation analysis with time varying exposures and mediators[J]. J R Stat Soc Series B Stat Methodol, 2017, 79(3): 917-938. DOI: 10.1111/rssb.12194.
[8]	Lee H, Herbert RD, McAuley JH. Mediation Analysis[J]. JAMA, 2019, 321(7): 697-698. DOI: 10.1001/jama.2018.21973.
[9]	方杰, 温忠麟, 张敏强, 等. 基于结构方程模型的多层中介效应分析[J]. 心理科学进展, 2014, 22(3): 530-539. DOI: 10.3724/SP.J.1042.2014.00530. Fang J, Wen ZL, Zhang MQ, et al. Analyzing multilevel mediation using multilevel structural equation models[J]. Adv Psychol Sci, 2014, 22(3): 530-539. DOI: 10.3724/SP.J.1042.2014.00530.
[10]	Stamps JA, Frankenhuis WE. Bayesian models of development[J]. Trends Ecol Evol, 2016, 31(4): 260-268. DOI: 10.1016/j.tree.2016.01.012.
[11]	Sint K, Rosenheck R, Lin H. Latent class mediator for multiple indicators of mediation[J]. Stat Med, 2021, 40(12): 2800-2820. DOI: 10.1002/sim.8929.
[12]	方杰, 温忠麟. 三类多层中介效应分析方法比较[J]. 心理科学, 2018, 41(4): 962-967. DOI: 10.16719/j.cnki.1671-6981.20180430. Fang J, Wen ZL. A comparison of three methods for testing multilevel mediation[J]. J Psychol Sci, 2018, 41(4): 962-967. DOI: 10.16719/j.cnki.1671-6981.20180430.
[13]	Rijnhart JJM, Lamp SJ, Valente MJ, et al. Mediation analysis methods used in observational research: a scoping review and recommendations[J]. BMC Med Res Methodol, 2021, 21(1): 226. DOI: 10.1186/s12874-021-01426-3.
[14]	侯杰泰, 温忠麟, 成子娟. 结构方程模型及其应用[M]. 北京: 教育科学出版社, 2004: 32-38. Hou JT, Wen ZL, Cheng ZJ. Structural equation model and its application[M]. Beijing: Education Science Press, 2004: 32-38.
[15]	徐洪吕, 伍晓艳, 陶舒曼, 等. 大学生饮料消费睡眠质量和抑郁症状的关系[J]. 中国学校卫生, 2020, 41(1): 16-20. DOI: 10.16835/j.cnki.1000-9817.2020.01.005. Xu HL, Wu XY, Tao SM, et al. Beverages consumption, sleep quality and depressive symptoms in Chinese university students: a latent variable mediation model[J]. Chin J Sch Health, 2020, 41(1): 16-20. DOI: 10.16835/j.cnki.1000-9817.2020.01.005.
[16]	Vrublevska J, Trapencieris M, Rancans E. Adaptation and validation of the Patient Health Questionnaire-9 to evaluate major depression in a primary care sample in Latvia[J]. Nord J Psychiatry, 2018, 72(2): 112-118. DOI: 10.1080/08039488.2017.1397191.
[17]	Ju M, Tao Y, Lu Y, et al. Evaluation of sleep quality in adolescent patients with osteosarcoma using Pittsburgh Sleep Quality Index[J]. Eur J Cancer Care (Engl), 2019, 28(4): e13065. DOI: 10.1111/ecc.13065.
[18]	Du Y, Du J, Liu X, et al. Multiple-to-multiple path analysis model[J]. PLoS One, 2021, 16(3): e0247722. DOI: 10.1371/journal.pone.0247722.
[19]	方绮雯, 刘振球, 袁黄波, 等. 结构方程模型的构建及AMOS软件实现[J]. 中国卫生统计, 2018, 35(6): 958-960. https://www.cnki.com.cn/Article/CJFDTOTAL-ZGWT201806045.htm Fang QW, Liu ZQ, Yuan HB, et al. Construction of structural equation model and realization of AMOS software[J]. Chin J Health Statistics, 2018, 35(6): 958-960. https://www.cnki.com.cn/Article/CJFDTOTAL-ZGWT201806045.htm
[20]	侯杰泰, 温忠麟, 成子娟. 结构方程模型及其应用[M]. 北京: 教育科学出版社, 2004: 43-48. Hou JT, Wen ZL, Cheng ZJ. Structural equation model and its application[M]. Beijing: Education Science Press, 2004: 43-48.
[21]	王卫东. 结构方程模型原理与应用[M]. 北京: 中国人民大学出版社, 2010: 17-24. Wang WD. Principles and applications of structural equation models[M]. Beijing: Renmin University of China Press, 2010: 17-24.
[22]	王孟成. 潜变量建模与Mplus的应用(基础篇)[M]. 重庆: 重庆大学出版社, 2014: 205-207. Wang MC. Latent variable modeling and the application of mplus (the basics)[M]. Chongqing: Chongqing University Press, 2014: 205-207.
[23]	王孟成. 潜变量建模与Mplus的应用(进阶篇)[M]. 重庆: 重庆大学出版社, 2018: 291-293. Wang MC. Latent variable modeling and the application of Mplus (advanced)[M]. Chongqing: Chongqing University Press, 2018: 291-293.
[24]	Rindskopf D. Overview of Bayesian statistics[J]. Eval Rev, 2020, 44(4): 225-237. DOI: 10.1177/0193841X19895623.
[25]	Miočević M, MacKinnon DP, Levy R. Power in Bayesian mediation analysis for small sample research[J]. Struct Equ Modeling. 2017, 24(5): 666-683. DOI: 10.1080/10705511.2017.1312407.
[26]	Miočević M, Golchi S. Bayesian mediation analysis with power prior distributions[J]. Multivariate Behav Res, 2022, 57(6): 978-993. DOI: 10.1080/00273171.2021.1935202.
[27]	Baron RM, Kenny DA. The moderator-mediator variable distinction in social psychological research: conceptual, strategic, and statistical considerations[J]. J Pers Soc Psychol, 1986, 51(6): 1173-1182. DOI: 10.1037//0022-3514.51.6.1173.
[28]	Zhao X, Lynch JG, Chen Q. Reconsidering baron and Kenny: myths and truths about mediation analysis[J]. J Consum Res, 2010, 37(2): 197-206. DOI: 10.1086/651257.
[29]	Enders CK, Fairchild AJ, Mackinnon DP. A Bayesian approach for estimating mediation effects with missing data[J]. Multivariate Behav Res, 2013, 48(3): 340-369. DOI: 10.1080/00273171.2013.784862.
[30]	Williams J, Mackinnon DP. Resampling and distribution of the product methods for testing indirect effects in complex models[J]. Struct Equ Modeling, 2008, 15(1): 23-51. DOI: 10.1080/10705510701758166.
[31]	Mackinnon DP, Lockwood CM, Williams J. Confidence limits for the indirect effect: distribution of the product and resampling methods[J]. Multivariate Behav Res, 2004, 39(1): 99. DOI: 10.1207/s15327906mbr3901_4.
[32]	方杰, 张敏强. 中介效应的点估计和区间估计: 乘积分布法、非参数Bootstrap和MCMC法[J]. 心理学报, 2012, 44(10): 1408-1420. DOI: 10.3724/SP.J.1041.2012.01408. Fang J, Zhang MQ. Assessing point and interval estimation for the mediating effect: distribution of the product, nonparametric Bootstrap and Markov chain Monte Carlo Methods[J]. Acta Psychol Sin, 2012, 44(10): 1408-1420. DOI: 10.3724/SP.J.1041.2012.01408.
[33]	Lefebvre G, Samoilenko M, Boucoiran I, et al. A Bayesian finite mixture of bivariate regression model for causal mediation analyses[J]. Stat Med, 2018, 37(25): 3637-3660. DOI: 10.1002/sim.7835.
[34]	Huang J, Yuan Y. Bayesian dynamic mediation analysis[J]. Psychol Methods, 2017, 22(4): 667-686. DOI: 10.1037/met0000073.
[35]	Miocevic M, Mackinnon DP, Levy R. Power in Bayesian mediation analysis for small sample research[J]. Struct Equ Modeling, 2017, 24(5): 666-683. DOI: 10.1080/10705511.2017.1312407.
[36]	Sakai H, Murakami K, Kobayashi S, et al. Food-based diet quality score in relation to depressive symptoms in young and middle-aged Japanese women[J]. Br J Nutr, 2017, 117(12): 1674-1681. DOI: 10.1017/S0007114517001581.
[37]	Roberts RE, Duong HT. The prospective association between sleep deprivation and depression among adolescents[J]. Sleep, 2014, 37(2): 239-244. DOI: 10.5665/sleep.3388.
[38]	Rindskopf D. Reporting Bayesian results[J]. Eval Rev, 2020, 44(4): 354-375. DOI: 10.1177/0193841X20977619.
[39]	陶秋山, 詹思延, 李立明. 流行病学研究中的病因与病因推断[J]. 中华流行病学杂志, 2004, 25(11): 86-89. DOI: 10.3760/j.issn:0254-6450.2004.11.021. Tao QS, Zhan SY, Li LM. Etiology and etiological inference in epidemiological research[J]. Chin J Epidemiol, 2004, 25(11): 86-89. DOI: 10.3760/j.issn:0254-6450.2004.11.021.