Can literature | relationship research draw causal conclusions

Hello, Hello everyone ～

This is one brain cloud research circle , I'm Xiao Hu ～

“ The greater the grip strength , The longer you live .”

“ The younger you feel, the healthier you are .”

When you see such a title , Do you have a question ： Is it wrong to regard correlation as cause and effect ？

In the comment area of the research on the relationship between subjective age and health level forwarded by domestic popular science we media , It can be seen that some netizens doubt the cause and effect deduced from the Correlation ：

Picture source ： video “ Research says the younger you feel, the healthier you are ” Comment area

Such questioning is not unreasonable , Then whether the results of this study are really what you netizens say , There is a reverse influence relationship ？

Browse this article published in geriatrics Feeling Younger, Rehabilitating Better: Reciprocal and Mediating Effects between Subjective Age and Functional Independence in Osteoporotic Fracture and Stroke Patients (Kalir et al., 2022), We will find that the author measured at the two time points of admission and discharge , The relationship between variables is derived through the cross lag model .

that , Whether the cross lag model can deduce causality in relevant research ？

Cross lag model , Also known as causal model 、 Cross lag panel model 、 Autoregressive cross lag model , It is generally used to discuss the influence relationship of reciprocating among multivariable ( Liu Yuan , 2021).

stay 7 Age and 17 At the age of two, the preference for violent TV programs and the tendency of aggressive behavior were measured

The basic cross lag model is usually in 2 Time point pair 2 Measure variables . As shown in the figure above , The key of the cross lag model is the cross lag relationship , The box “7 I prefer violent TV programs at the age of ” forecast “17 Year old aggressive behavior tendency ”, Or with “7 Aggressive behavior tendency at the age of ” forecast “17 I prefer violent TV programs at the age of ”.

Inferring causality through cross lag relationship upholds this assumption ： Cause precedes effect , That is, the time sequence can explain the causal relationship to a certain extent . This assumption assumes that the variables measured later will not in turn have an impact on the previously measured variables (Shingles, 1976).

therefore , If found in the model above “7 I prefer violent TV programs at the age of ” Be able to predict “17 Aggressive behavior tendency at the age of ”, and “7 Aggressive behavior tendency at the age of ” Can't predict “17 I prefer violent TV programs at the age of ”, It can be inferred that the preference of watching violent TV programs affects the tendency of aggressive behavior , Instead of the opposite direction of influence .

Because it is based on “ Cause precedes effect ” Assumptions , The reliability of cross lag inference causality is sensitive to the measurement time interval . Building a cross lag model requires that the measurement interval be consistent with the actual “ reason - result ” Lag matching . Take the basic cross lag model as an example , Specifically, there are usually several principles to follow (Shingles, 1976)：

1. The first 1 The time taken for this measurement must be shorter than the actual “ reason - result ” Lag interval , In order to avoid mixing the cross lag relationship between variables in the cross-sectional relationship ;

2. The first 2 This measurement must wait until the actual lag is completely over , However, it cannot be delayed until the causal effect is significantly weakened with the passage of time , Maximize the causal relationship between variables .

Take the relationship between sedative dosage and anxiety as an example

therefore , The cross lag method is used to explore the relationship between variables , We need to estimate the duration of the effect in advance to get a more accurate amount of effect .

Even if the measured time point coincides with the actual lag time , There are still inherent limitations in inferring causality with cross lag relationship ：

1. Due to the failure to control irrelevant variables , The relationship between variables may still be caused by other irrelevant variables ( Liu Yuan , 2021);

2. The cross lag relationship is a mixture of intra subject variation of individual development and variation of different development laws between individuals , As a result, inferring causality through cross lag relationship will overestimate the amount of effect (Berry & Willoughby, 2017);

3. The relationship between variables inferred by the cross lag model based on data is often difficult to explain with the causal relationship assumed in advance (Berry & Willoughby, 2017). The competition between several similar available models may also generate mixed causal conclusions ;

Because of these limitations , The inference of general cross lag model cannot be called causal relationship strictly , And is defined as “ Reciprocating effect ”( Liu Yuan , 2021).

To make a long story short , Although compared with the experimental method , The causal relationship inferred by the cross lag model is not so “ rigorous ”, But when discussing the causal relationship between variables that are difficult to manipulate in experiments , The inference of causality by cross lag model still provides us with more convincing evidence than simple cross-sectional research .

reference

[1]Berry, D., & Willoughby, M. T. (2017). On the practical interpretability of cross‐lagged panel models: Rethinking a developmental workhorse. Child development, 88(4), 1186-1206.
[2]Kalir, D. M., Shrira, A., Palgi, Y., Batz, C., Ben-Eliezer, A., Heyman, N., Lieberman, D., Seleznev, I., Shugaev, I., Shugaev, I., Zaslavsky, O., Zikrin, E., Bodner, E (2022). Feeling Younger, Rehabilitating Better: Reciprocal and Mediating Effects between Subjective Age and Functional Independence in Osteoporotic Fracture and Stroke Patients. Gerontology, 1-9.
[3]Shingles, R. D. (1976). Causal inference in cross-lagged panel analysis. Political Methodology, 3(1), 95-133.
[4] Liu Yuan . (2021). Model integration and expansion of multivariable tracking research : Review the impact and growth trend of reciprocating . Progress in Psychological Science , 29(10), 1755.

author ｜ Xiao Hu

Typesetting ｜Uka

proofreading ｜ Elder sister miaojun Kun Kun