On Monday 4th of Februrary (11am) we will have a journal club on google search data and its use for measuring mental health and how psychological distress reacts to economic events.
Background:
The degree to which economic fluctuations have substantive mental health effects remains an open question. The difficulty in addressing this issue often stems from inadequate data. Although there are an increasing number of large scale surveys gauging mental health and well-being levels of representative samples of the population, these are often conducted at an annual or less frequent basis. However, researchers in psychology, economics, and other disciplines have begun to study internet search terms as a method for measuring mental health at a very high frequency level. With many countries approaching 80% internet penetration and with the internet acting as often the initial source of information for people who feel troubled with mental health problems, this strategy seems one worth thinking about. There are many internet tools that have been used for health surveillance (listed here). In particular, the tool Google Trends captures the magnitude of search volume of particular terms over the period from 2004 onwards. It adjusts for changes in the overall volume of internet search and allows both time and geographical changes to be examined. This tool has been used to track flu epidemics and touted as a method for its sensitivity in capturing rapid changes in prevalent conditions (described here , current flu trends here).
Journal club topics:
The first topic for discussion in the journal club will be the evidence for the validity of google mental health searches (e.g. searching for 'symptoms of depression', 'depression test', 'stress symptoms') as measures of psychological distress.
Previously, Google Trends data has been used in a widely cited paper by Choi & Varian to measure changes in economic indicators like car sales, when people claim unemployment, and are planning to go abroad. In this instance, highly accurate data is available but lags behind search data due to the process of collection and validation. Search therefore provides a clearer picture of the present than the older economic data that is available at a given time. Whether this is the case in the instance of changes in mental health is an important question that has not been fully addressed to date.
Three useful papers that aim to address the validity of search data in its application to mental health have recently been publised in the Journal of Affective Disorders and will be the first paper reviewed in the journal club. They show that the volume of searches for terms like 'Major depression' and 'Suicide' appear to coincide with statistics for completed suicides over time, though the strength of the relation may vary by age and whether temporal or geographic patterns are examined. Though a study by Page and colleagues (2011) failed to identify this relationship.
McCarthy, M.J. (2010). Internet monitoring of suicide risk in the population. Journal of Affective Disorders, 122, 377-379.
Yang, A.C. et al. (2011). Association of Internet search trends with suicide death in Taipei City, Taiwan, 2004–2009. Journal of Affective Disorders, 132, 179-184.
A recent paper by Gun III and Lester examined the differences in suicide searches and rates across the 50 American states and identified evidence for correspondence.
Gun III, J.F., & Lester, D. (in press). Using google searches on the internet to monitor suicidal behavior. Journal of Affective Disorders.
The Journal of Affective Disorders appears to be at the forefront in examining the link between internet search volumes for psychological distress and how these match to objectively verified data. There is other evidence to suggest search data may reflect changes in mental health. For instance, Yang and colleagues (2010) have shown seasonal variation in depression searches to occur, particularly at higher latitudes. Media reporting of suicide has also been reported to have a lagged effect on suicide deaths (Yang et al., 2012).
There is now growing evidence that the financial crisis may have had adverse effects on mental health as indexed by increases in suicide levels in the EU (e.g. Stuckler et al., 2011) and countries such as Greece (e.g. Kentikelenis et al., 2011) and Italy (De Vogli et al., 2012). In the most recent instance a rise in Italian suicides with the cause 'due to economic reasons' was identified.
Given the potential linkages between internet search volumes for mental health terms and suicide rates it follows that these searches may act as a sensitive measure of the psychological effects of economic changes. Recent articles have addressed this idea and will be reviewed in the journal club along with the three articles referenced above:
Ayers, J.W. et al. (2012). Novel surveillance of psychological distress during the great recession. Journal of Affective Disorders, 142, 323-330.
Tefft, N. (2011). Insights on unemployment, unemployment insurance, and mental health. Journal of Health Economics, 30, 258-264.
We ask the following questions:
1) What are the benefits of using mental health search data over existing data sources (rapid, spontaneously generated, sensitive, less incentives to misreport, potentially sensitive to micro-level changes, can be mapped onto rich geographical data, potential for integration with existing data sources).
2) What are the potential confounding factors associated with search data, in this instance for examining mental health (selected sample- though large internet penetration in many nations, search terms may not reflect the inferred concept (e.g. 'depression' could be psycholgical or economic), social events and media coverage may cause peaks in search, currently not possible to profile searchers by age, gender, income and so forth but this is potentially solvable as, temporal changes may reflect the impact of third variables like the availability of internet resources).
3) How should mental health/psychological distress search term dependent variables be formed? (e.g. multiple dependent variables, composite, latent factor).
4) What would a strong validation test of psychological distress search terms look like?
5) What are the counterfactual tests that can be used to assist in validating analyses linking search data and economic events or trends. For instance, whilst 'depression symptom' searches increases in line with global financial trends so do 'cancer symptom' trends (see below). The former is feasibly an effect of the financial crisis whereas the latter is most likely not.
6) Should control search terms be used and how?
7) Should time series or geographical data be used or a combination? What are the likely pitfalls or each approach and the analytic techniques that may benefit the examination (e.g. fixed-effects)?
8) Are the effects of the financial crisis identified in the articles above consistent? Are they feasible in magnitude?
9) What are the potential policy applications of mental health search data?
10) How should we think about the relationship between search terms at the group level and individual mental health problems? How does this interact with statistical/logical problems such as the ecological fallacy?
Background:
The degree to which economic fluctuations have substantive mental health effects remains an open question. The difficulty in addressing this issue often stems from inadequate data. Although there are an increasing number of large scale surveys gauging mental health and well-being levels of representative samples of the population, these are often conducted at an annual or less frequent basis. However, researchers in psychology, economics, and other disciplines have begun to study internet search terms as a method for measuring mental health at a very high frequency level. With many countries approaching 80% internet penetration and with the internet acting as often the initial source of information for people who feel troubled with mental health problems, this strategy seems one worth thinking about. There are many internet tools that have been used for health surveillance (listed here). In particular, the tool Google Trends captures the magnitude of search volume of particular terms over the period from 2004 onwards. It adjusts for changes in the overall volume of internet search and allows both time and geographical changes to be examined. This tool has been used to track flu epidemics and touted as a method for its sensitivity in capturing rapid changes in prevalent conditions (described here , current flu trends here).
Journal club topics:
The first topic for discussion in the journal club will be the evidence for the validity of google mental health searches (e.g. searching for 'symptoms of depression', 'depression test', 'stress symptoms') as measures of psychological distress.
Previously, Google Trends data has been used in a widely cited paper by Choi & Varian to measure changes in economic indicators like car sales, when people claim unemployment, and are planning to go abroad. In this instance, highly accurate data is available but lags behind search data due to the process of collection and validation. Search therefore provides a clearer picture of the present than the older economic data that is available at a given time. Whether this is the case in the instance of changes in mental health is an important question that has not been fully addressed to date.
Three useful papers that aim to address the validity of search data in its application to mental health have recently been publised in the Journal of Affective Disorders and will be the first paper reviewed in the journal club. They show that the volume of searches for terms like 'Major depression' and 'Suicide' appear to coincide with statistics for completed suicides over time, though the strength of the relation may vary by age and whether temporal or geographic patterns are examined. Though a study by Page and colleagues (2011) failed to identify this relationship.
McCarthy, M.J. (2010). Internet monitoring of suicide risk in the population. Journal of Affective Disorders, 122, 377-379.
Yang, A.C. et al. (2011). Association of Internet search trends with suicide death in Taipei City, Taiwan, 2004–2009. Journal of Affective Disorders, 132, 179-184.
A recent paper by Gun III and Lester examined the differences in suicide searches and rates across the 50 American states and identified evidence for correspondence.
Gun III, J.F., & Lester, D. (in press). Using google searches on the internet to monitor suicidal behavior. Journal of Affective Disorders.
The Journal of Affective Disorders appears to be at the forefront in examining the link between internet search volumes for psychological distress and how these match to objectively verified data. There is other evidence to suggest search data may reflect changes in mental health. For instance, Yang and colleagues (2010) have shown seasonal variation in depression searches to occur, particularly at higher latitudes. Media reporting of suicide has also been reported to have a lagged effect on suicide deaths (Yang et al., 2012).
There is now growing evidence that the financial crisis may have had adverse effects on mental health as indexed by increases in suicide levels in the EU (e.g. Stuckler et al., 2011) and countries such as Greece (e.g. Kentikelenis et al., 2011) and Italy (De Vogli et al., 2012). In the most recent instance a rise in Italian suicides with the cause 'due to economic reasons' was identified.
Given the potential linkages between internet search volumes for mental health terms and suicide rates it follows that these searches may act as a sensitive measure of the psychological effects of economic changes. Recent articles have addressed this idea and will be reviewed in the journal club along with the three articles referenced above:
Ayers, J.W. et al. (2012). Novel surveillance of psychological distress during the great recession. Journal of Affective Disorders, 142, 323-330.
Tefft, N. (2011). Insights on unemployment, unemployment insurance, and mental health. Journal of Health Economics, 30, 258-264.
We ask the following questions:
1) What are the benefits of using mental health search data over existing data sources (rapid, spontaneously generated, sensitive, less incentives to misreport, potentially sensitive to micro-level changes, can be mapped onto rich geographical data, potential for integration with existing data sources).
2) What are the potential confounding factors associated with search data, in this instance for examining mental health (selected sample- though large internet penetration in many nations, search terms may not reflect the inferred concept (e.g. 'depression' could be psycholgical or economic), social events and media coverage may cause peaks in search, currently not possible to profile searchers by age, gender, income and so forth but this is potentially solvable as, temporal changes may reflect the impact of third variables like the availability of internet resources).
3) How should mental health/psychological distress search term dependent variables be formed? (e.g. multiple dependent variables, composite, latent factor).
4) What would a strong validation test of psychological distress search terms look like?
5) What are the counterfactual tests that can be used to assist in validating analyses linking search data and economic events or trends. For instance, whilst 'depression symptom' searches increases in line with global financial trends so do 'cancer symptom' trends (see below). The former is feasibly an effect of the financial crisis whereas the latter is most likely not.
6) Should control search terms be used and how?
7) Should time series or geographical data be used or a combination? What are the likely pitfalls or each approach and the analytic techniques that may benefit the examination (e.g. fixed-effects)?
8) Are the effects of the financial crisis identified in the articles above consistent? Are they feasible in magnitude?
9) What are the potential policy applications of mental health search data?
10) How should we think about the relationship between search terms at the group level and individual mental health problems? How does this interact with statistical/logical problems such as the ecological fallacy?
No comments:
Post a Comment