Multilayer regression analysis and psychological intervention effect assessment study of college students’ tennis intervention for depressed mood

A questionnaire survey was conducted in an undergraduate college between May and June 2018 using the whole cluster sampling method according to the principle of informed consent. A total of 270 college students were surveyed, and 270 valid questionnaires were obtained, of which 80 male students, 120 female students, and 70 non-participants participated in tennis exercise, with an average age of (20.13±1.19) years.

A self-developed questionnaire was used to conduct the survey, which included general demographic information, tennis playing status, and mental health status.

EpiData 3.0 was used to input the data and SPSS 13.0 software was used to analyze it statistically. Between-group comparisons of the current status of tennis sports and the detection rate of depressive symptoms were performed using the x² test. Multilayer regression analysis was performed with depressive symptoms as the dependent variable and general demographic information and tennis sport status as independent variables [19].

The research method in this paper is multilevel regression modeling. Although the impact of community neighborhood effects on individual poverty can also be analyzed using ordinary OLS regression, the data in the paper involves different levels (including the individual, household and community levels), and there is a correlation between the residual terms of individuals in the same community, and the direct use of OIS regression is likely to result in the inaccuracy of model measurements, and the multilayer regression model allows for the existence of a correlation between observations, which, in turn, can improve the estimation results accuracy. In the process of constructing the multilevel regression model, it is possible to distinguish the effects of different levels and groups of variables on the explanatory variables. The estimation methods of multilevel regression models include both great likelihood estimation and restricted great likelihood estimation, which can alleviate the problems of omitted variables and data imbalance that may occur during the modeling process. The multilevel regression model in this paper contains three levels, which are set as follows:

Level-1 (individual level): 1Pr(Yijk=1|eijk)=H(π0jk+πijkXijk+π2jkX2ijk+⋯+πpjkXpjk+eijk)eijk~N(0,σ2)$$\matrix{ {\Pr ({Y_{ijk}}\> = \>1\>|{e_{ijk}}) = H\left( {{\pi _{0jk}}\> + \>{\pi _{ijk}}\>{X_{ijk}}\> + \>{\pi _{2jk}}\>{X_{2ijk}}\> + \> \cdots \> + \>{\pi _{pjk}}\>{X_{pjk}}\> + \>{e_{ijk}}} \right)} \hfill \cr {{e_{ijk}}\unicode {x223C}N\left( {0,{\sigma ^2}} \right)} \hfill \cr } $$

Level-2 (family level): 2πpjk=γpjk+γpjkZ1jk+γpjkZ2jk+⋯+γpjkZ0jk+upjkupjk~N(0,σpjk2)p=0,1,2,⋯,P$$\matrix{ {{\pi _{pjk}} = {\gamma _{pjk}} + {\gamma _{pjk}}{Z_{1jk}} + {\gamma _{pjk}}{Z_{2jk}} + \cdots + {\gamma _{pjk}}{Z_{0jk}} + {u_{pjk}}} \hfill \cr {{u_{pjk}}\unicode {x223C}N\left( {0,\sigma _{pjk}^2} \right)} \hfill \cr {p = 0,1,2, \cdots ,P} \hfill \cr } $$

Level-3 (community level): 3γpqk=δpq+δpqlW1k+δpq2W2k+⋯+δpqsWsk+upqkupqk~N(0,σpqk2)q=0,1,2,⋯,Q$$\matrix{ {{\gamma _{pqk}} = {\delta _{pq}} + {\delta _{pql}}{W_{1k}} + {\delta _{pq2}}{W_{2k}} + \cdots + {\delta _{pqs}}{W_{sk}} + {u_{pqk}}} \hfill \cr {{u_{pqk}}\unicode {x223C}N\left( {0,\sigma _{pqk}^2} \right)} \hfill \cr {q = 0\>,\>1\>,\>2\>,\> \cdots ,\>Q} \hfill \cr } $$

Of these: Y_ijk is the poverty status of the individual, X_pijk is the individual level variables, including the health status, age, gender, and education level of the individual. Z_pk is household level variables, including the class the household is in, net household income, and household size. W_sk are community-level variables, including urbanization index, community education category, population density, and transportation accessibility. Neighborhood effect variables are also community-level variables, including community poverty rate, average community education, and average community income.

Of course, the multilevel regression model can be used only when the explanatory variables differ significantly between groups. ICC [20] is a common indicator to determine whether the data are applicable to the multilevel regression model. When the ICC tends to 0, it means that there is no significant between-group difference in the explanatory variables and the data are not applicable to the multilevel regression model; when the ICC tends to 1, it indicates that there is a significant between-group difference in the explanatory variables and the data are applicable to the multilevel regression model. As long as 0.059 < ICC, the between-group differences are not negligible and a multilevel regression model should be constructed. The ICC indicator needs to be calculated according to the fully unconditional/null model as follows:

Null model setting: 4Pr(Yijk=1|eijk)=H(π0jk+eijk) 5π0jk=γ00k+u0jk 6γ00k=δ000+u00k 7Yijk=δ000+eijk+u0jk+u00k

Therefore, the variance of the explanatory variables is: 8Var(Yijk)=Var(δ000+eijk+u0jk+u00k)=σ2+σ0jk2+σ00k2

Calculation of ICC indicators: 9ICCCommunity=σ00k2σ2+σ0jk2+σ00k2 10ICCFamily=σ0jk2+σ00k2σ2+σ0jk2+σ00k2

This paper uses data from the China Health and Nutrition Survey (CHNS) for 1991-2011 [21]. Although the survey is not specifically designed for poverty-related research, it contains information on individual income, household information, and community information that offer possibilities for poverty research. In order to retain as many observations as possible, this paper organizes the data for all the years covered by the survey, selects the key variables, deletes the missing values, and retains the samples with age greater than or equal to 18 years old, leaving a final sample of 80,000.

Individual-level variables were selected to include education level, health status, gender, age, and age squared; household-level variables were selected to include social class, net household income, and household size; and community-level variables were selected to include urbanization index, type of education in the community, population density, and accessibility to transportation in the community. Community neighborhood effects variables include the community’s poverty rate, average education level, and average income. Note that the neighborhood effects in this paper exclude individual i’s own circumstances. Descriptive statistical information for each variable is shown in Table 1. Most individuals in the sample are not poor, but the number of poor people under the WB2$PPP poverty criterion is much higher. Most individuals have elementary school education and above, and are 43 years old or older. Most households are located in urban and rural areas, with the majority of households having four or above bedrooms.

Interest in extracurricular tennis was 48% among students, followed by 17% who could not say, 17% who were less interested, 12% who were very interested, and 6% who were not interested respectively.

Intensity of Tennis Exercise 34% of the students chose micro-exercise and 33% small-intensity less intense exercise, followed by 19% large-intensity but not long-lasting exercise, 13% medium-intensity more intense long-lasting exercise, and 1% large-intensity continuous long-lasting exercise, respectively.

The frequency of tennis exercise was chosen by the students who exercised 2 to 3 times per month in the first place, about 33%, followed by 1 to 2 times per week with 30%, less than 1 time per month with 22%, 3 to 5 times per week with 115 and 1 time per day respectively 45

The time of tennis exercise was chosen by 28% of the students who exercised 31-59 min each time, followed by 25% of 21-30 min each time, 24% of 11-20 min each time, 16% of more than 60 min each time and 7% of less than 10 min each time, respectively.

The evaluation of tennis exercise effect was done by using the self-assessment scale of college students’ tennis exercise effect (EEI), and the overall mean ± standard deviation of the five dimensions and the comparison between male and female students are shown in Table 2. As shown in Table 2, there were differences between male and female students’ satisfaction in the dimensions of exercise enjoyment (t=2.64, P=0.02) and socialization effect (t=2.72, P=0.02), and male students’ satisfaction was generally higher.

The SDS scale is rated by the frequency of symptoms and is divided into 4 levels: none or very little of the time, a little of the time, quite a lot of the time, and most or all of the time. If the questions were positively scored, they were rated as crude scores l, 2, 3, and 4, and if negatively scored, they were rated as 4, 3, 2, and 1. The scores of the 20 questions were summed up as the crude score, which was multiplied by 1.25, and rounded up to the nearest whole number to obtain the standardized score. The cutoff value in the Chinese norm was 53 points, with 53-62 points being mild depression, 63-72 points being moderate depression, and 72 points or more being severe depression.

In this study, 24 college students showed different degrees of depression, with an incidence rate of 12%. Of these, 19 were mildly depressed, 4 were moderately depressed, and 1 was severely depressed.

Using SPSS13.0, the linear correlation analysis was conducted between the scores of the dimensions of tennis sport effect and the SDS standard score, and the results are shown in Table 3. It can be seen that there is a negative correlation between the scores of the dimensions of tennis sports effect and the total score and the standard score of SDS scale, except for the dimension of appearance effect, and the p-value of the overall score of tennis sports is 0, which is less than 0.01, and the difference is statistically significant. It can be concluded that with the improvement of tennis sports, the depression condition of college students has been improved.

The multilevel regression analysis affecting depression status was performed using individual level, family level, community level, exercise interest, exercise intensity, exercise frequency, exercise time and exercise effect evaluation as the influencing factors, and the results are shown in Table 4. It can be seen that the difference between those who played tennis for less than 20 minutes per session and those who played for more than 60 minutes was statistically significant [OR (95% CI) = 3.15 (1.14-6.63), P = 0.03]. The difference between those who were very dissatisfied with the results of tennis exercise and those who were very satisfied [OR (95% CI)=4.12 (1.54-11.13), P=0.01], and those who were dissatisfied and those who were very satisfied [OR (95% CI)=3.12 (1.18-8.16), P=0.03] was statistically significant. It can be concluded that with the improvement of the time spent on each tennis session and the effect of tennis exercise, the depression status of the students has improved accordingly.