The problem with retrospective reports: change

When multiple measures are taken over intervals of time, change in those measures can be assessed. For example, pain intensity measures assessed before, during and after an intervention can be examined for change. However, when multiple measures are not available, researchers instead rely on patients to supply information on change by asking, for example, "how much has your pain improved since you began the treatment last month?"

The problem is that past experience is not recorded in memory as images are recorded in a video. People use a host of mental shortcuts (heuristics) to reconstruct the past. One of these heuristics is to use current experience as a guide to the past. If there is no reason to believe that the past should be any different from the now then the now becomes the past and that's what is reported. On the other hand, if there is some reason to believe the past my be different from the past then reports will be adjusted accordingly. For example, if people undergo a treatment for pain, their expectation of improvement will lead people to assume that their current pain should be less than their past pain. That is, pain should be improved after treatment compared to before treatment. According to Schwartz (2010), "asking patients whether they feel better now than before their treatment is the most efficient way to "improve" the success rate of medical interventions."

The problem with retrospective reports: intensity

The problem with asking people about the intensity (How painful? How much anxiety? How optimistic? etc.) of a past experience is that intensity is not very well represented in memory. Therefore, when asked about intensity, people must use various mental tricks to reconstruct an answer. The problem is that these reconstructions of past experience bear little resemblance to the past experience as it was experienced.

In a frequently cited study, Redelmeier and Kahneman (1996) asked patients undergoing either colonoscopy or lithotripsy to rate their pain every 60 sec by moving a marker on a computer screen to indicate pain intensity. Then while recovering, within one hour of their procedure, patients were asked to assess their "total amount of pain experienced". Attending physicians were also asked to judge the overall pain experience of each patient. The researchers found that patients' retrospective reports of overall pain experienced were strongly correlated with both peak pain (.64) and end pain (.44) during the final 3 mins of the procedure. They also found that longer procedures were not predictive of recalling greater amount of pain. Even though a longer procedure subjects people to more pain overall, people don't seem to take account of the entire experience in their retrospective reports. Instead, people tend to recall only how bad it got (peak), and how it ended (end).

In addition to the peak/end heuristic, people also resort to various inference strategies. For example, asked about last week's pain, patients often try to contract an answer by noting how much pain they currently have, and then consider whether last week's pain might have been different. If so, the pain report will reflect this anomaly (Ross & Conway, 1986, Ross, 1989; Linton & Melin, 1982), otherwise the current pain will assumed to be representative of last week's pain (Eich et al, 1985) and that's what will be reported. As Schwartz (2010) put it, "... her retrospective report of pain is a function of her current pain and her naive theory about the stability of her pain over time."

What is ecological momentary assessment (EMA)?

According to Arthur Stone, who is one of the most prolific researchers and authors of EMA methodology, EMA refers to the "repeated collection of real-time data on participants' momentary states in the natural environment," and that "the key elements of EMA are real-time collection of data about momentary states, collected in the natural environment, with multiple repeated assessments over time." (Stone, Shiffman, Atienza, Nebeling, 2010). Let's briefly look at each of these elements.

Real-time data collection: Data are captured about individuals as life happens.

This means that instead of asking people how they felt or what they did in the past (which is much more frequently the case in health research), we capture how they feel as they're feeling it, and what they're doing, as they're doing it.

If a doctor asks you how much pain you've been having over the past month, you will likely try to recall how high your pain got and what proportion of the time your pain was at this level. This process puts a great burden on memory, which a tremendous volume of research has shown is fallible and subject to numerous biases that can shift the reality that we remember quite far from the reality that actually occurred.

EMA avoids this reliance on memory altogether by capturing data about mental and physical states at the time those states are occurring.

Momentary states: What's happening in and to individuals at brief slices of time (i.e., moments)

Imagine an assembly line making computers. To ensure quality, a robot is programmed to pluck individual computers from the line at random intervals and runs tests on them. You can think of each computer as a moment and the results of the tests as the data regarding the state of the system (the assembly line). That is data are collected on the behavior of the system by measuring the state of the computers at moments in time. Notice that not every computer (i.e., moment) is tested. Sure, testing every computer would provide an extremely detailed picture of the system but it would also pose a very heavy burden to do so. Therefore computers (moments) are sampled from among all possible computers (moments).

In the same way, EMA samples moments in peoples' lives. The data that are collected -- the variables -- depends on the research questions. Participants can be asked to rate how they currently feel on various emotions (happy, excited, frustrated, etc.), about the location and intensity of their physical pain, and their current level of physical capability on some scale, say, from 0 (not at all) to 10 (very much). Now, increasingly capable yet inexpensive devices are appearing that enable the measurement of a host of physiological parameters such as heart rate, skin conductance (for sympathetic arousal), heart rate variability, blood pressure, electromyography, and more. Things get particularly exciting when the relationship between self-report measures are correlated against objective physiological measures (but that's a topic beyond the scope of this article).

Natural environment: Data capture occurs as people go about their lives.

This is where the "ecological" in EMA comes in. Instead of bringing people into a lab, doing something to them, and then measuring their responses, we follow people out "in the wild" in their "natural habitats", where all the complexities of life and all the forces acting on people are preserved.

As an example, for my doctoral dissertation I investigated the effects of social disconnectedness on physical pain. In one of my studies, healthy undergrads were invited into the lab where they were exposed to a social interaction with a partner who was in fact a collaborator of the researchers posing as another participant. For some participants (the lucky ones), the partner was very warm and friendly, whereas for other participants, the partner was cool and aloof. Physical pain sensitivity was measured both before and after this social exchange. Everything was scripted and tightly controlled. But the problem is that uses contrived experiences in a relatively safe setting (a university lab) that are isolated from all the wonderful complexity of life. And this should make us wonder whether the results we obtain in such experiments apply beyond the walls of the lab, which is really the place where we hope the results apply because after all, life doesn't happen in a lab, it happens "out there". EMA techniques and technologies allow us to get "out there".

Repeated assessments over time: Data are captured multiple times daily over many days.

In most experimental studies, measurements are taken once. Participants roll in, measures are taken, then participants roll back out. In pre-post studies, measures are taken twice. In follow-up studies, measures may be taken three or four times. These studies are typically referred to, collectively, as repeated measures (RM) studies. EMA should not be confused with RM studies. EMA studies typically involve multiple assessments per day, repeated over a number of days. This dense sampling provides a level of temporal resolution that permits the examination of how dynamic processes unfold over time. For example, we can look at whether certain physiological parameters (arousal) immediately precedes anxious thoughts, or whether increasing levels of pain vs. steadily high pain levels lead to differences in psychological wellbeing.

Most devices that track heart rate data require a cheststrap. Speaking from personal experience, they're uncomfortable. If you strap them too loosely they fall slip down when you get sweaty and then fail to record heart rate accurately. To keep it from sliding you have to strap them quite snugly, which is just not too comfortable. The Scosche myTrek is a $130 device that you wear on an armband that senses and sends heart rate data wirelessly (via bluetooth) to an iPhone or iPod touch. The iPhone/iPod app provides a nice interface that you can use to watch heart rate in one of 5 training zones, calories burned, distance/speed, workout duration and overall progress towards goals that you pre-specify.

Scosche myTrek: http://www.scosche.com/mytrek

SenseWear device can measure activity, energy expenditure and sleep

Worn on an armband, the BodyMedia SenseWear device contains a number of sensors: accelerometer to measure activity levels and steps taken, galvanic skin response (the only highly mobile, relatively inexpensive, device that I've seen that does this), skin temperature, and "heat flux" (amount of heat dissipating from the body). It can hold 28 days' worth of data and the raw data can be downloaded in Excel format. So it can be used to measure energy expenditure, step count, time spent in varying levels of activity (sedentary, moderate, vigorous), and sleep parameters (time lying down, sleep duration, sleep efficiency).

They also have analysis and visualization software. Each device is $500 but a sales rep told me they offer quantity discounts: for quantities of 10-19 armbands, you get a 10% discount off list. If you order > 20, then you receive a 20% discount and the next discount applies for quantities of 50 or more. A bit expensive but it measures a lot. This product is promoted to clinicians and so is probably reliable; plus it has a great number of professional abilities and tools (e.g., you can configure channels for higher or lower sampling rates, etc.).

BodyMedia SenseWear: http://sensewear.bodymedia.com

AliveCor iPhone based ECG

This device (http://alivecor.com) looks like any case that you'd snap onto an iPhone, but it's got 2 stainless steel sensor plates on the back. To capture ECG you just put his/her left and right thumbs on the left and right plates (or place the whole device on the chest) and you get an instant clean ECG stream.

Check out http://alivecor.com/video.htm for demos from the inventor.

It's still awaiting FDA clearance.

Shimmer wireless sensor system

Cardio Development Kit

Shimmer (http://www.shimmer-research.com) makes several small, wearable, wireless sensors that can measure, store, and transmit a number of physiological (ECG, EMG, GSR) and kinematic (accelerometer, gyro, vibration, GPS) data in real-time. Data are stored in the sensors and then can be wirelessly transferred to whatever devices you want (iPhone, iPad, laptop computer, etc.).

The cool thing about this system is that it is completely configurable and programmable. You can capture whatever phys data you want. They provide the wireless sensors and programming tools but you decide how you want to use them. You get a programmer who can configure the devices to capture exactly the data you need, at whatever schedule you want, and then do what you want to the raw data (could be just formatting it in a text file or Excel file format). They also have a LabView module.

For some ideas of applications see the following 2 pages:

- Phys applications: http://www.shimmer-research.com/applications-2/biophysical
- Kinematic applications: http://www.shimmer-research.com/applications-2/kinematics

They have a number of starter kits that they sell.

Most of my research has been in the area of pain and the kinematics stuff is really cool for pain apps because it enables the assessment of objective behavioral measures of treatment effectiveness. For example, you could detect actual degree/range of movement (using gyro to detect angle, location in space).

Striiv Device Aims to Make Fitness Fun Through Feedback

Striiv is a small device that tracks your steps. But any pedometer can do that. What makes Striiv interesting is what it does with the information it captures. Steps are turned into points to unlock achievements, that you can use to build things, grow plants and create buildings. And the more you create, the more coins you earn to build more things. It also has a very intriguing feature that enables you to turn steps into donations. The company Striiv has teamed with charities and donors so that walking counts towards donations that are made on your behalf (for no cost to you -- just walking). Check it out at http://www.striiv.com/

Study predicts illness with cellphone usage records

I recently came across a paper called "Social Sensing for Epidemiological Behavior Change". It describes a study in which smartphone use of college students was used to identify behavior changes associated with the presence of illness (colds, flu, fever, stress, depression).

The researchers gave smartphones to college students in a dormitory and tracked phone usage with call data records, sms logs, proximity sensing, and location-based sensing. Students also completed a brief questionnaire regarding the presence of symptoms. Proximity sensing enabled detecting when these student participants were physically close to other participants. The researchers found that student who developed a fever or cold tended to move around less and made fewer calls in the morning and late at night.

I think this is absolutely fascinating. It makes me think that there may be a whole world of meaning that can be extracted from data that are readily available but that we're missing. I rarely hear anyone in my field, Psychology/Neuroscience, talk about pattern recognition technologies -- we're trained to formulate hypotheses (expectations based upon what we know). But this is limiting. There's a world of surprises awaiting us if we just know how to look.

Track sleep unobtrusively with under-the-mattress sensor

A new device developed by BAM Labs, uses an inflatable pad that is placed under a mattress to unobtrusively track heart rate, respiration, motion and presence. Then data collected are wirelessly transmitted to a cloud-based service on the Internet, which can then be accessed through a website or with mobile devices.

This looks like a great tool for researchers. The really nice thing is that it tracks sleep parameters without requiring that individuals wear or attach anything to the body making it more acceptable to research subjects. Also, researchers (and caregivers) can monitor and collect data from many individuals remotely.

Unfortunately, it isn't yet available nor have they published release dates or prices.

Track heart rate without cheststrap

Most devices that track heart rate data require a cheststrap. Speaking from personal experience, they're uncomfortable. If you strap them too loosely they fall slip down when you get sweaty and then fail to record heart rate accurately. To keep it from sliding you have to strap them quite snugly, which is just not too comfortable. The Scosche myTrek is a $130 device that you wear on an armband that senses and sends heart rate data wirelessly (via bluetooth) to an iPhone or iPod touch. The iPhone/iPod app provides a nice interface that you can use to watch heart rate in one of 5 training zones, calories burned, distance/speed, workout duration and overall progress towards goals that you pre-specify.

Life tracking for clinical psychopharmacology

Life tracking techniques can provide a reliable means by which patients symptoms, thoughts, feelings and behaviors can be recorded in vivo, as they go about their daily lives.

Moskowitz and Young (2006) examined clinical psychopharmacological studies to see what methods were being used to measure affect and social behavior.

They found that the primary method reported was clinician judgment based on first-hand observation or information provided by the patient to the clinician. The second most commonly used method was patient-completed questionnaires.

Twice as many studies examined affect as measured social behavior. Measurements of affect were typically quite detailed with the use of multiple instruments assessing different aspects, whereas measurement of social behavior, was far more coarse-grained, and did not assess a wide range of important factors such as the quality of the social interactions in which patients were involved, characteristic behaviors in social situations, characteristics of others or the context that might have an impact on the individual's symptoms. As the authors put it, "Despite the centrality of social behavior to descriptions of psychopathology, social behavior is examined less frequently and in less detail in psychopharmacological studies than mood and affect." (p. 14).

Clinician reports

Clinician reports are the most common used method in clinical psychopharmacology to investigate social behavior and affect.

The problems with clinician reports:

1. The accuracy of clinician assessments can be affected by familiarity with the individual yet in most studies, clinicians have little contact with participants and so are not at all familiar with the individuals they're assessing. As a result, clinicians rely heavily on info provided by the participant.
• Moskowitz and Young (2006) point to a classic study by Rosenhan (1973) to demonstrate the questionable validity of clinician assessments. In that study, it was found that although clinicians couldn't detect pseudo patients admitted to psychiatric wards, other patients (who spent more time with them) were able to detect who was who.
2. Reliability between multiple clinicians is typically not assessed. When there is an attempt to establish inter-rater reliability, it typically takes the form of different raters coding the same taped clinical interview. Unfortunately, this practice does not establish the the consistency with information is elicited from participants across clinicians.
3. Clinicians have pre-existing assumptions of how various measures co-vary.
• Schachar et al (1986) found that in teachers, defiant behavior in a student influences teachers' evaluations of hyperactivity and inattentiveness (which are believed to be correlated with defiance) such that defiance toward a teacher increases the likelihood that a child will be rated as hyperactive and inattentive, even in the absence of actual observations of hyperactivity and inattentiveness.

Self-report questionnaires

The problems with questionnaires:

1. It is difficult to control for all types of response biases. For example, people who are highly neurotic may be prone to respond more negatively. and to recall symptoms.
2. Responses are affected by one's mood at the time of assessment. Thus assessments made at particular days or times may heavily reflect the timing of the assessment.
3. Recalled information is subject to reconstructive processes.

A method of tracking change that does not rely on clinical assessments or memory would thus prove very useful.

Ecological Momentary Assessment

1. Collect info at pre-specified time intervals. Can be once or several times a day.
2. Signal-contingent recording
1. Reports triggered by a signal that occurs randomly for some fixed number of times per day.
2. All participants given the same number of signals and so report on the same number of events.
3. Events important to the research may be missed.
3. Event-contingent recording
1. Reports triggered by the occurrence of some event.

One of the great advantages of EMA is its ability to investigate changes over time.

• Can look at consistency of affect and behavior over time.
• Can look at the lability of affect and behavior in response to particular events.
• Given enough data, can look at the sequence of things -- whether change in affect might precede or come after changes in behavior.
• The relationship between variables within subjects and the way by which pharmacological agents might impact this relationship can be examined.
• e.g., in depressed patients, the sequence of improvement in mood and improvements in social interactions (duration, agreeableness) can be examined.

"Day Reconstruction Method" offers way to track diurnal affect rhythms

The Day Reconstruction Method (DRM) was designed to enable the tracking of activities from the previous day while minimizing recall biases. Participants are given the following instructions...

"Think of your day as a continuous series of scenes or episodes in a film. Give each episode a brief name that will help you remember it (e.g., “commuting to work” or “at lunch with B”). Write down the approximate times at which each episode began and ended. The episodes people identify usually last between 15 minutes and 2 hours. Indications of the end of an episode might be going to a different location, ending one activity and starting another, or a change in the people you are interacting with."

For each episode listed, participants select what they were doing (from a provided list), with whom they were interacting (if anyone), and select from among a list of 12 emotion adjectives to indicate how they were feeling.

Stone et al (2006) had 909 women come in large groups to complete the DRM.

Findings regarding the DRM...

1. Mean num of episodes per day was 14 and median episode length was 61 mins.
2. Were able to complete the exercise in less than 1 hour.
3. Able to recruit and run large numbers of participants.
4. Authors point out that experiences can be expanded to include other variables such as symptoms and health behaviors.

Findings regarding diurnal emotion cycles...

1. Bimodal pattern observed for both positive and negative emotions.
2. For the 3 positive emotions, intensity had a 1st peak at around noon and a 2nd peak in the evening.
3. For the negative emotions, intensity peaked around 10 a.m., and then again at 4 or 5pm (although this pattern was not observed for all negative emotions).
4. The emotion tired did not fit the bimodal pattern. It conformed to a "v" shaped pattern. Tired reached its lowest point around noon and then increased as the day progressed. The authors speculate that this is because tired is independent of daily activities; that it is more dependent of physiological processes than the other emotions.
5. Diurnal cycles observed for all emotions, though some were more strongly tied to time of day than others.
• Strongest diurnal pattern observed for tired was the emotion most strongly linked to time of day. Time of day explained > 18.5% of the variance in the tired.
• Weakest diurnal patterns occurred for criticized, depressed, and angry.
6. Overall, positive affect (happy, enjoy) increases throughout the day, whereas negative feelings (angry, depressed, frustrated, worry) decrease. Tired decreases throughout the day.
7. Particularly interesting was that after statistically accounting for activities the diurnal patterns for enjoy and frustrated flattened out. For example, enjoy showed a peak at noon; this peak was eliminated after partialling out the effect of activities (e.g., lunch).

This research was limited to females so it remains to be seen whether males will exhibit diurnal emotion cycles similar or distinct from females.

---

Stone, A. A., Schwartz, J. E., Schkade, D., Schwarz, N., Krueger, A., & Kahneman, D. (2006). A population approach to the study of emotion: Diurnal rhythms of a working day examined with the day reconstruction method. Emotion, 6(1), 139–149. doi:10.1037/1528-3542.6.1.139

Study of college students finds no link between time spent in physically sedentary and active behaviors

Rouse and Biddle (2010) asked British university undergrads to record their main behavior, where they were and whom they were with, every 15 mins.

They found that the top activities were (in descending order): studying, shopping/hanging out, tv viewing, computer use, and sitting talking. Males spent significantly more time studying than females.

Most interesting, out of a total of 558 hours (for females), only 63 were spent doing things that involved physical activity -- that's only 11%!!

Another interesting finding is the lack of a correlation between "technological sedentary behaviors" (tv, computer, video games) and physical activity, suggesting that one has little to do with the other. As the authors point out, this finding runs counter to popular perception and suggest that efforts to increase physical activity by reducing sedentary behavior may be met with limited success.