Edited by: Sven Braeutigam, University of Oxford, UK
Reviewed by: Christian Lambert, St. George's University of London, UK; Jessica Clare Scaife, Oxford Univeristy, UK
*Correspondence: Martin Block, Northwestern University, Medill Integrated Marketing Communications, MTC 3-123, 1845 Sheridan Road, Evanston, IL 60208, USA e-mail:
This article was submitted to the journal Frontiers in Human Neuroscience.
†,‡Authors made equal contributions, corresponding to First (†) or Second (‡) authorship.
This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.
Depression is a debilitating condition that adversely affects many aspects of a person's life and general health. Earlier work has supported the idea that there may be a relationship between the use of certain media and depression. In this study, we tested if self-report of depression (SRD), which is not a clinically based diagnosis, was associated with increased internet, television, and social media usage by using data collected in the Media Behavior and Influence Study (MBIS) database (
Depression is known to affect many kinds of human behavior, and is quite common. As of 2005, the lifetime prevalence of major depressive disorder in the US population was reported to be 16.5% (Kessler et al.,
There is a developing literature evaluating the relationship between various types of media use and psychiatric conditions. For instance, one study found a high positive correlation between internet addiction and depression among university students (Orsal et al.,
This study differed from previous studies in the following ways. (1) The sample size of the dataset was substantially larger than any previous study evaluating the relationship between media use and depression. (2) We evaluated the link between depression and multiple domains of media use, whereas most previous studies have focused primarily on single domains. For example, recent work with a smaller database has suggested there is an increase in digital media usage in “depressed” adolescents (Primack et al.,
Our analysis started with descriptive and bivariate statistical analyses. These were followed by omnibus approaches to assess general effects given the number of variables describing media usage: (a) Chi-squared Automatic Interaction Detection or CHAID tree analysis (Kass,
The dataset was derived from the Media Behavior and Influence Study (MBIS), a syndicated online study of American adult (i.e., ≥18 years of age) consumers, conducted twice yearly since 2002 by BIGinsight of Columbus, Ohio. The current wave of 19,776 participants was completed in December, 2012. Using a double opt-in methodology, each MBIS study was balanced to meet demographic criteria established by the US census. MBIS data has been used by a variety of well-known, commercial marketing organizations. Variables of interest included depression by gender, age, employment status, marital status, race and ethnicity, income, education, measures of isolation, and internet, TV and social media use. These variables were selected because they have been variables of interest in previous depression studies, and have been shown to have predictive value (e.g., Catalano and Dooley,
Three types of analysis were performed with this data. First, we performed a descriptive statistical analysis, inclusive of correlations between depression and media consumption variables to facilitate interpretation of the subsequent analyses. Second, we used the results to inform a type of recursive partitioning (Morgan and Sonquist,
We performed two recursive partitioning analyses, one focused on SRD and the second on a variable not of interest, namely non-SRD, to act as a control for SRD results. Our working hypothesis was that the control analysis of non-SRD subjects would not replicate or provide an opponent (i.e., completely non-overlapping) set of nodes to the analysis of SRD subjects.
Construction of statistical CHAID trees (SPSS tree) evaluated the interaction among a number of predictor variables of SRD, and separately non-SRD. Typically, such schemes are defined in terms of demographic variables such as age and gender; however we have also included occupation, education, marital status and media use. Splitting criteria included minimum parent node size of 100 and child node size of 50, and a
Discriminant analysis was used to conduct a multivariate analysis of variance for the hypothesis that people who self-reported having depression would differ significantly from non-SRD subjects on a linear combination of eleven variables: income, internet usage, TV usage, social media usage, education, age, living in top 10 metropolitan area (MSA), gender, having children, employment status, and disability. The discriminant analysis was run using SPSS defaults, resulting in the canonical linear discriminant analysis. Depression was the binary dependent variable entered in the “group” dialog. The discriminating variables were entered together (i.e., not stepwise) in the variables subcommand. The discriminating variables income, internet usage, TV usage, social media usage, and education, all took on continuous values in the range from 0 to 1. “Living in top 10 MSA,” gender, employment status and disability were binary categorical variables while having children was ordinal. Overall, the data were complete with no missing values (i.e., every subject had every data point).
The MBIS study shows little to no geographic pattern for SRD (Figure
The data does show that SRD among all adults in the USA (18 and over) has grown from 11.2% in 2009 to 12.1% in December, 2012, with a linear trend (
Age (average years)* | 45.4 | 43.8 | 96.3 |
Male | 48.3 | 11.8 | 97.5 |
Female | 51.7 | 12.3 | 101.7 |
Income (000)* | 62.8 | 49.0 | 78.0 |
Have children | 29.1 | 30.2 | 103.9 |
Live in top 10 MSA | 24.7 | 10.0 | 82.6 |
Married | 42.5 | 9.5 | 78.5 |
Living with unmarried partner | 7.2 | 15.5 | 128.1 |
Divorced or separated | 10.2 | 15.4 | 128.3 |
Widowed | 3.0 | 12.4 | 102.5 |
Single, never married | 25.7 | 14.1 | 116.5 |
Same sex union | 0.5 | 22.2 | 183.5 |
Have not graduated high school | 1.5 | 21.7 | 179.3 |
Graduated high school | 16.8 | 13.1 | 108.3 |
Technical school or vocational training | 5.7 | 13.8 | 114.0 |
1–3 years of college (did not graduate) | 20.2 | 15.1 | 124.8 |
Associates or professional degree | 8.9 | 13.2 | 109.1 |
Bachelor's degree | 22.5 | 9.4 | 77.7 |
Post college study or degree | 13.5 | 8.8 | 72.7 |
Business Owner | 4.2 | 11.7 | 96.7 |
Professional/managerial | 25.5 | 8.2 | 67.8 |
Salesperson | 3.6 | 11.5 | 95.0 |
Factory worker/laborer/driver | 3.3 | 9.6 | 79.3 |
Clerical or service worker | 9.5 | 11.9 | 98.3 |
Homemaker | 3.6 | 14.7 | 121.5 |
Student, high school or college | 8.4 | 13.0 | 107.4 |
Military | 0.7 | 11.6 | 95.9 |
Retired | 13.7 | 10.8 | 89.3 |
Unemployed | 5.5 | 18.8 | 155.4 |
Disabled (unable to work) | 2.0 | 42.7 | 352.9 |
Obsessive-compulsive disorder (OCD) | 2.1 | 9.8 | 458.1 |
Anxiety | 12.9 | 54.8 | 425.8 |
Dyslexia | 0.8 | 2.6 | 334.5 |
Fibromyalgia | 2.3 | 7.4 | 327.5 |
Insomnia/difficulty sleeping | 8.4 | 27.5 | 325.8 |
Restless leg syndrome(RLS) | 4.1 | 11.6 | 279.8 |
Irritable Bowel Syndrome (IBS)/crohn's disease | 2.4 | 6.3 | 262.6 |
Chronic bronchitis/COPD | 2.9 | 7.3 | 252.5 |
Sleep apnea | 6.6 | 16.4 | 248.2 |
Heartburn/indigestion | 9.9 | 22.7 | 230.3 |
Headaches/migraines | 14.0 | 29.6 | 211.5 |
Back pain | 21.5 | 42.7 | 198.4 |
Acid reflux | 15.8 | 30.5 | 192.9 |
Heart disease | 3.2 | 5.9 | 185.9 |
Hearing impairment | 4.3 | 8.0 | 184.6 |
Overweight | 21.2 | 37.6 | 177.1 |
Arthritis | 15.5 | 27.3 | 176.4 |
Asthma | 9.7 | 17.0 | 174.8 |
Vision impairment | 15.0 | 24.8 | 165.2 |
Enlarged prostate/Benign Prostatic Hyperplasia (BPH) | 2.2 | 3.6 | 162.6 |
Diabetes | 9.3 | 15.0 | 162.1 |
Osteoporosis | 2.5 | 4.1 | 161.6 |
High cholesterol | 18.8 | 29.3 | 155.9 |
Black | 18.0 | 8.7 | 71.9 |
Asian | 3.0 | 7.9 | 65.3 |
Multi | 0.8 | 16.9 | 139.7 |
Native | 0.4 | 15.5 | 128.1 |
White | 58.4 | 13.6 | 112.4 |
Other | 0.5 | 9.9 | 81.8 |
Hispanic | 18.9 | 10.9 | 90.1 |
Media usage quintiles, a method commonly used in the media industry, were created using the composite media usage variables described above, and showed higher rates of depression among the most active users of media. Figure
It should be noted that there was some co-linearity between the three media categories. The correlation of television and internet consumption was moderate at 0.495, slightly higher for internet and social media at 0.510, but lower for television and social media at 0.247. All of these correlations were significant (
The analyses reported above were limited to bivariate correlations. To better understand how multiple variables for media consumption and other demographics/activities related to SRD, a multivariate segmentation scheme was employed based on recursive partitioning (Morgan and Sonquist,
In the analysis of SRD subjects, the CHAID tree segments (Figure
In the analysis of non-depressed individuals (non-SRD), the CHAID tree segments (Figure
It is important to note that the CHAID analysis with non-SRD did not replicate the analysis with SRD. Furthermore, there was a segmentation observed between these analyses which was distinct, in that the types of media use that segmented the SRD subjects was not the same as that which segmented the non-SRD subjects. The terminal nodes of the two analyses were different along dimensions of occupation, income, and media use.
The results of the discriminant analysis revealed that, other than disability and income, the three single best predictors of depression in this model were increased use of television, the internet, and social media (Table
Disabled | 0.760 |
Income | −0.519 |
Internet usage | 0.399 |
TV usage | 0.368 |
Social media usage | 0.278 |
Education | −0.255 |
Unemployed | 0.223 |
Age | −0.170 |
Living in top 10 MSA | −0.142 |
Female | 0.062 |
Having children | 0.010 |
The primary finding of this study is that those who tend to use more media in general, also tend to have more self-reports of depression. We found a current incidence of SRD at 12.1% which is slightly less than reports of lifetime clinical depression and more than the 12 month incidence of diagnoses of major depression. However, the picture is far more nuanced than simple description of descriptive statistics and bivariate correlations between media use and depression. For instance, the CHAID tree analysis with SRD subjects (along with the discriminant analysis) shows that those who have suffered either economic or physical life setbacks are orders of magnitude more likely to be depressed, even without disproportionately high levels of media use (37.2%). However, among those that have suffered major life setbacks, high media users—particularly television watchers—were even more likely to report experiencing depression (47.3% in the highest two quintiles, as compared with 35.2% in the lower three quintiles), which suggests that these effects were not just due to individuals having more time for media consumption. These effects were not observed with the control analysis in non-SRD subjects. That the economically disadvantaged are significantly more likely to experience depression is well supported by research in social psychology, which suggests that lower-income groups feel a sense of disempowerment (Henry,
Life challenges may not be the only experiences related to depression. As noted with our descriptive statistical analysis, persistent environmental factors such as isolation can also contribute to the prevalence of a psychological experience. Generally, isolation is a known correlate of depression symptomology, and our data suggest that residents of rural areas tend to report higher rates of depression. Within the context of isolation, one can distinguish between physical and non-physical isolation; and within non-physical isolation one can look at social and emotional isolation. These various subclasses of isolation find ample support in the literature. Weiss (
In addition to the current state of depression, the data we analyzed reveals that SRD has been in a state of flux over the past decade. At the beginning of this time frame, the rates we observed were low compared to 2005 MBIS data where the depressive rate was reported to be 14.9%; a figure consistent with a co-occurring 2005 study wherein a lifetime prevalence rate of 16.5% for major depressive disorder was reported (Kessler et al.,
It is worth considering the demographics of individuals (e.g., gender) reporting SRD in the context of a flux in depression rates over time. As recently as 5 years ago, females were more likely to report being depressed (i.e., SRD). However, in the most recent MBIS study, the data shows SRD to be similarly associated with both genders, with males reporting only a slightly lower rate of depression. This is different than the rates reported by Primack et al. (
There are several important limitations to this study that are worth mentioning. First, the data used was self-reported depression, which does not necessarily reflect whether the subject has ever received a clinical diagnosis of depression. The subjective phenotypes of those who have a clinical diagnosis of major depression versus those that self-report depression could skew the data in a number of different ways. For instance, it has been observed that those who have been diagnosed with depression are sometimes reticent to share their diagnosis. Alternatively, there is a multiplicity of reasons to think that subjects without depression may report being depressed. The balance of these considerations leaves uncertainty in the true sample parameters, although the percentage of subjects with SRD in this study was quite similar to rates of depression found in previous studies.
Second, the variables computed for amount of television, internet, and social media use are not direct measures. These variables are composite variables computed from self-reports of whether or not subjects used those various media during discrete variable-hour-length blocks. This can introduce inter-subject variability along a number of dimensions. For instance, some subjects may report “yes” for one of the intervals based on an hour's worth of use, while others may respond the same based on several hours' worth of use. The probabilities computed represent just that, a probability of time spent using a given media relative to other subjects.
Third, the analyses done cannot speak to a causal relationship between media consumption and depression, or to any directionality between the observed associations. We think the likeliest explanation is that these two variables form a complex bi-directional relationship with autocatalytic properties. An alternative explanation is that depression and increased media use are a byproduct of a third confounding factor. It should also be noted that the direction of causality between depression and media use could also vary across individuals (i.e., whether media usage helps to ameliorate depression or whether it contributes to it). Whatever the exact relationship between depression and increased media use, it is clear that the two are closely associated.
Fourth, it is important to acknowledge the potential confounds of concurrent medical illness on assessing associations with SRD. In the literature on major depression, hypotheses have been raised that depression in association with a medical illness does not necessarily reflect the same structural and functional circuitry alterations seen in depression with strong familial heritability (e.g., see Cloninger,
This information can help to form hypotheses to test in future studies of relevance to psychology. One such hypothesis could relate to the directionality of the relationship between SRD and media, to determine if any media use acts as feedback to exacerbate symptoms. Another hypothesis might attempt to relate the relationship to existing social psychological constructs such as the “empty self” hypothesis. Cushman (
In summary, the data reveal that there is a consistent pattern of results that link self-reported depression with increased media use, even when taking into account other variables, such as disability and unemployment. This media use was focused more on internet use and TV exposure, for those making self-reports of depression. The rate of SRD was between two standard indices used in published reports of clinically diagnosed major depression, namely the lifetime prevalence, and recent 12 month incidence of major depression. These observations suggest the current findings with big data may have relevance to the literature focused on the clinical diagnosis of depression.
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
The Supplementary Material for this article can be found online at: