Helping behavior during COVID-19 lockdown

This brief research project was part of the course “Linear models” in the Master in Methodology for behavioral sciences.

Description

Why we help other people is a question psychologists have been trying to answer for decades. Researchers usually design experiments in which participants have the opportunity to help. However, a non-experimental approach can also shed some light on the issue.

Here I adopt that perspective and make use of a public survey conducted by the spanish Center for Sociological Research (CIS) after the end of the 2020 lockdown. Because the outcome of interest, helping behavior, is described in the survey as a dichotomic variable (either you help or you don’t), I decide to fit a logistic model of the form:

$$ \begin{aligned} \hat P(help_i=\text{Yes}|{\bf x}_i)=\frac{1}{1+e^{-(\beta_0+{\boldsymbol{\beta}^\intercal\cdot {\bf x}_i})}} \end{aligned} $$

where ${\bf x}_i$ represents the vector of $J$ features or explanatory variables of the $i$-th subject. Read the equation as: the estimated probability of helping for person $i$, given a set of features that person has, is modeled as a (logistic) function of a linear combination of those features.

The model specification is partially based on Schwartz and Howard (1984). They describe five stages in helping behavior:

1. attention to need,

2. generation of value-based motivation,

3. awareness of potential actions,

4. evaluation of the costs and benefits of potential actions, and

5. overt behavior or inaction.

However, I don’t contemplate values as the main motives for helping behavior but empathy. The features finally selected were:

• Type of people affected (attention),

• Purposes of change (motivation),

• Feeling of being helpful (awareness/self-efficacy),

• Grief over loss of loved one (empathy),

• Concern over loss of job (empathy) and

• Sex.

Full description of them can be found here.

Summary of results

• Cross-validated Nagelkerke’s $R^2$ is approximately 0.07. The proposed model reduces the misfit of the null model by 7%.

• From the point of view of the model’s classification performance, the results are also modest, with an AUC value of 0.62.

• The profile of people with the lowest propensity to help is as follows: men who consider that the crisis has mainly affected the people who have suffered it directly, have not suffered from the loss of family, friends or employment, have not made any intention to change, and do not consider their actions to be particularly helpful. The prognosis is an odds ratio of 0.468, which implies that the probability of helping is 53.2% lower than the probability of not helping.

• The indirect measures of empathy are the best predictors of helping behavior. Holding all else constant, the odds of helping is 1.715 times greater among individuals who report having suffered from the loss of a family member or friend than among those who do not report having suffered this experience. This data is consistent with that found in the sample, in which the proportion (in percentage) of people who say they help and have suffered from the loss of people is 15 points higher than that of people who say they help but have not suffered from the loss of loved ones.

The complete report can be found here.