Me and the Other Guy: A Data-Driven View of a Chaotic Mind
I analyzed years of my own digital footprint seeking clues to my Bipolar disorder. I found more than I expected.
Disclaimer: This essay contains personal experiences and data analysis related to bipolar disorder. It is not intended as medical advice. Always consult with healthcare professionals for medical decisions. Just want the guide? Join here!
It was afternoon on an average weekday that I realized I was in trouble. I knew the place, but I didn’t know how I’d gotten there, and I certainly didn’t understand why my phone was making so much noise. My lips were numb. I was rambling at length to strangers. I had at least three napkin business plans, ten new ideas for essays, and a rock solid conviction that I was finally in full control of my life.
I had also stood up eight clients at my practice, trapped an inexperienced employee in a situation she shouldn’t have been in, and totaled my car.
I’d been diagnosed with Bipolar I early in life, but hadn’t ever really accepted it. I was first treated as a teenager, but the symptoms faded through my 20s and early 30s and I gradually came to believe it wasn’t true. Around 35 it hit me like a truck, dragging me along for a disruptive journey that I’d rather be able to predict or avoid.
In early 2023 I decided to take a big data approach to find strategies. I started publicly journaling my experiences on Bluesky Social when the platform launched, which began as an exercise in record-keeping and disability awareness but has led to something else: an incredibly useful written record of my actions and emotions with feedback from a wonderful community of people who have experienced the same.
I’m writing this to share the process, my experience during the experiment, and what I’ve learned — which is quite a lot.
9/7/2023
I wish mental health disorders were seen by everyone as the invisible disability that they are. I had someone last week tell me that I shouldn’t write about my bipolar here where I’m identifiable and I’m like … if you had an arm chopped off in your 30s do you think it would ever come up? @coolhand
To establish bona fides. By sheer coincidence of life experiences, I’m in a somewhat unique position: As an Applied Linguist, my Masters thesis was a computational analysis of the speech of people diagnosed with Schizophrenia. I’ve worked as a Speech-Language Pathologist specialized in complex (read: severe) communication and thought disorders. I’ve watched a family member live with an acute mental health condition — and now I’m watching myself live with my own. Finally, I’m a seasoned product person and software engineer — of dubious skill in the latter, but with a history of creating small language models and analyzing complex biometrics for health trends.
None of these things are particularly unique or impressive, and I don’t share them to appear that way. I share them because my background is almost comically on the nose, and because I don’t want this content to be misunderstood. The data shared and visualized is authentic, but simplified in most cases and obfuscated in others. X and Y axes may lack labels when focused on trends, and while I’ll release the process I won’t release my own personal data (for what I hope are obvious reasons). Similarly, nothing shared or learned here should be taken as clinical advice; this is an essay, not a journal article, and lacks citations to prove it.
Finally, while I’ll speak to privacy, hallucination, and other issues with AI, they deserve more space and consideration on this topic than I can afford them. From the fact that I’m sharing this essay, you might infer that privacy is not my top priority. Even outside of that, use of AI is rightfully considered to be problematic in a wide variety of ways. That said, I believe that disability and accessibility are one of the most obvious use cases — and that the tools that make many apprehensive can, for others, make things possible.
The data used for my analysis includes:
- Amazon and other online shopping receipts
- Apple Health data (steps, sleep)
- BlueSky Social: posts (also my primary journal)
- Facebook: posts, media, location history, and messages
- Google: Gmail, location history, and calendar events
- iMessage history
- LinkedIn: posts, messages, and employment history
- WhatsApp and Telegram message histories
- YouTube and Vimeo views and uploads
- Weather, major news and world events
- Writing and other content creation (Medium, Substack, Patreon)
- Other content such as family biographies and potential emotional trigger words to validate
Selected findings are below, organized by each bipolar state and ending with what I’ve found of potential predictors.
1. Euthymia
n. [Modern Latin from Greek, from eu ‘well’ + thymos ‘soul, emotion’]
A state of psychological well-being characterized by contentment, emotional stability, and a positive outlook on life.
I was diagnosed with bipolar at eleven. That’s a little earlier than most; in fact, there’s a standing joke in the community that the pipeline to diagnosis leads through ADHD and Autism Spectrum Disorder before landing on the truth. Technically my original diagnosis was manic depression, now properly known as cyclothymia, but evolved rapidly from there. It remained a manageable condition for many years, and one that I took with more than a little salt; a few extra sick days here, a few extra long emails there. Around 35 the experience intensified considerably, which isn’t too uncommon for so-called “formal thought disorders;” it’s the outside edge for initial diagnosis, but right in the sweet spot of where things can get worse.
Euthymia isn’t a “normal” mood state in the way most people would conceptualize. It’s characterized simply by the absence of mania or depression, a liminal space of clarity that can last weeks or months or a day. While it never seems stable or long-lived, it does serve as a baseline by which the rest can be judged. It represents a goal state of relative normalcy that can be achieved and maintained with effort and awareness.
From the standpoint of raw activity across all data sources, euthymia (for me) sits solidly between depression and mania. It’s what my friends would consider “regular” me. I’m physically active, spend a fair bit of time on social media, don’t overspend (usually), and maintain a consistent sleep schedule. I tend not to use language that’s associated with negative people or events, and am generally a bit above neutral in terms of sentiment. It’s also the state where I get the most anxious, especially in regard to what state might be coming next.
10/29/2023
FR I’m super anxious today. I have like six weeks of stability which is unheard of for me so waiting for the other shoe to drop @coolhand
How long these periods last is highly individual and, at least to some extent, circumstance based. One of the most challenging things about bipolar is differentiating between chemically abnormal mood states and just … moods. Normal moods that everyone feels. In analyzing shifts in and out of euthymia, speed and severity tend to indicate the intensity of the following bipolar event. As someone considered a “rapid cycler,” those shifts often happen every few days. With all integrated data sources and mood journals to check against, it’s not difficult to identify them.
While Euthymic:
Sleep Patterns:
- Awake 72% of the day
- Consistent and restful sleep schedule
Emotional State:
- Maintain a consistent mildly positive communication tone
- Balanced emotional responses to situations
- Ability to experience and express a full range of emotions appropriately
Communication:
- Balanced communication across personal and professional platforms
- Appropriate frequency and content of social media posts
- Ability to engage in meaningful conversations without over-sharing
Stress Management:
- Share negative sentiments, but primarily related to normal stress
- Ability to cope with and process stressors effectively
Physical Activity:
- Maintain a reasonable and consistent exercise rhythm
- Regular running routine, typically several miles per day
- Balanced approach to physical health, without obsession or neglect
Productivity:
- Steady work output
- Ability to start and complete projects
- Realistic goal-setting and achievement
Self-Care:
- Consistent attention to personal needs and health
- Regular engagement in hobbies and relaxation activities
- Maintenance of social connections and support systems
The clearest precipitators of an escalation out of euthymia, for me, are a marked increase in social media posts and direct messages such as texts. I tend to check apps that I otherwise rarely use, such as Discord or Telegram.
Terminology questions? Download this free graphic for definitions!
Online purchases are also an indicator, although more so in later states, as is GitHub activity. I start using much more positive language and words related to action. Location history becomes more variable as I visit new places. My physical activity in terms of step counts tends to peak and then fall as I start to spend more time hyperfocused.
My physical state as measured by blood pressure … well, the numbers aren’t good. If all of those things increase, it’s a fair bet that the next step will be hypomania.
2. Hypomania
n. [Modern Latin from Greek, hypo- ‘under, below’ + mania ‘madness, frenzy’]
Persistent and pervasive elevated or irritable mood, increased energy and activity, and often an exaggerated sense of well-being.
People tend to think of bipolar in terms of mania and depression, possibly because of the old “manic depressive” diagnosis. There’s another place — hypomania — which is an escalated state that can be seductive in its perceived positivity. For Bipolar II, it’s considered the “peak” of an episode, often followed by a longer depression. For my diagnosis, Bipolar I, it’s an on ramp; I experience hypomania about three times more often than the negative equivalent — which is positive emotionally, but risky in other ways. This is the period, for example, when medication tends to slip.
12/2/2023
49–50 hrs w/o sleep now and I’m going for a run bc I have so much energy. I love this part of bipolar @coolhand
Looking at a six-month period with all datasets, there are clear consistent peaks above baseline — and occasional extreme events. Looking only at sentiment, the data lines up well with the overall analysis — but not perfectly. Negative terms are used all across the timeline, even when not associated with a manic event. Conversely, when depressed there’s less sentiment expressed overall.
Looking only at emotion in language, the data lines up well with the overall analysis — but not perfectly. Negative terms are used all across the timeline, even when not associated with a manic event. Conversely, when depressed there’s less emotion expressed overall.
While Hypomanic:
Sleep Patterns:
- Awake 88% of the time
- Reduced need for sleep without feeling tired
Emotional State:
- Consistently more positive sentiment overall
- Rare instances of negative sentiment
- Potential for irritability, especially when faced with obstacles
- Mood variations still present, but less extreme than in mania
Communication:
- Increased communication across all platforms
- Emphasis on professional networks (e.g., LinkedIn)
- Rapid and frequent messaging with assertive language
- Higher engagement in social interactions
Physical Activity:
- Higher levels of physical activity overall
- Erratic patterns in exercise routines
- Restlessness and difficulty sitting still
Productivity:
- Increased work output and creativity
- Tendency to start multiple new projects
- Improved problem-solving abilities
Social Behavior:
- Increased sociability and outgoingness
- Potential for risky or impulsive social decisions
Self-Care:
- Variable attention to personal needs
- Potential for neglect of routine health practices due to increased activity
- Visit more new destinations and range further on location tracking.
A hypothetical association I explored was periods of no activity whatsoever; twelve hour absences (usually due to sleep) almost universally precipitate a flip in sentiment, usually from positive to negative as a hypomanic or manic event resolves to depression.
Hypomania is a double-edged sword. On one side, it brings a surge of creativity, productivity, and social engagement. On the other, it carries the risk of impulsivity, poor decision-making, and the eventual crash into depression. During hypomanic episodes I’m inspired to start a hundred new projects, only to then never finish. This inconsistency can be frustrating and lead to a cycle of unfinished endeavors and strained relationships. Worst of all, it can lead to mania.
3. Mania
n. [Modern Latin from Greek, from mania ‘madness, frenzy’]
Extreme elevation of mood, increased energy and activity, inflated self-esteem, and significantly reduced need for sleep.
Mania is the extreme end of the bipolar spectrum, often misunderstood as simply excessive energy or happiness. While it can start with feelings of euphoria and boundless energy, it can quickly spiral into dangerous territory. Mania can lead to impulsive decisions, risky behaviors, and even psychosis.
During manic episodes, my digital footprint explodes. Financial transactions increase, which can lead to uncomfortable moments later down the line. My physical activity, as tracked by step counts and location data, becomes erratic and drops overall as I tend to enter technology-oriented hyperfocus. Sleep, or the lack thereof, is a critical marker — extended periods without sleep often signal the onset of mania, and certainly continue through mania itself.
8/27/2023
I’m burning you guys. I haven’t slept in almost 70 hours … everything just feels kind of thin to me right now. I’m fine don’t worry, I’ve been very open on here about bipolar. But this is a rough one even for me. @coolhand
Mania is both exhilarating and terrifying; it can feel like a superpower, but it comes at a cost. Relationships strain under the weight of unpredictable behavior, and the crash into depression is almost inevitable. Financial recklessness, strained relationships, and legal troubles are just a few of the potential outcomes. The aftermath of a manic episode often leaves me swamped with regret and the task of repairing the damage done — which means it’s by far the most important for me to predict.
While Manic:
Sleep Patterns:
- Awake 100% of the day, often for 3–4 consecutive days
- Extreme reduction in sleep need, sometimes complete insomnia
Communication and Social Media:
- Peak communication across personal messaging platforms
- Overwhelming volume of posts on specific social media sites
- Rapid and frequent messaging with more assertive or aggressive language
- Wide variation in communication partners
Productivity and Creativity:
- Significant increase in coding activity (observable via GitHub history)
- Initiation of multiple new projects (e.g., domain name purchases)
- Thoughts and communications often lack coherence or logical flow
Emotional State:
- Highly variable sentiment, rapidly shifting between euphoria and irritation
- Inflated self-esteem and sense of capabilities
Physical Activity and Location:
- Generally decreased physical activity due to hyperfocus
- Brief, intense spikes in activity (e.g., sudden urge to go running)
- Wide variation in locations visited
Financial Behavior:
- Substantial increase in purchases, often impulsive or unnecessary
- Investment in new project ideas, regardless of practicality
Self-Care:
- Neglect of basic needs (food, water, general health and safety)
- Reduced awareness of physical limitations or consequences
10/9/2023
Bipolar event escalating I can feel it, ozone smell is back and my lips numb and I’m talking constantly. I hope it isn’t true but I think mania real soon. Updated charts for my own ref in next analysis. @coolhand
While it isn’t universally the case, it can generally be assumed that what goes up must come down; after mania comes depression.
4. Depression
n. [Modern Latin from Latin, from deprimere ‘press down’]
Persistent low mood, decreased energy and activity, negative thoughts, and a diminished sense of well-being.
For me, depression is a deep well where the world is muted and nothing interests me at all. Fortunately this is uncommon for me, and doesn’t generally last very long, but it’s the sort of state where finding food feels like Olympic gymnastics.
12/10/2023
it appears the poles have swapped. I was just starting to enjoy the mania.
During depressive episodes, my social media presence nearly disappears. Communication with friends and family becomes limited to essential interactions. Physical activity plummets, as tracked by health data. Sleep patterns become erratic, with either excessive sleeping or insomnia. On the plus side, I don’t spend any money.
While Depressed:
Sleep Patterns:
- Awake only 31% of the day
- Excessive sleep or desire to sleep constantly
- Potential insomnia despite fatigue
Communication and Social Interaction:
- Nearly complete cessation of 1:1 communication, with rare exceptions
- Significant disengagement from professional platforms
- Drastically reduced frequency of personal social media posts
Emotional State:
- Persistent negative sentiment
- Occasional breaks in negativity through self-deprecating humor
- Diminished sense of self-worth and hopelessness
Media Consumption:
- Passive consumption of media, often negative in nature
- Tendency towards “doomscrolling” or consuming anxiety-inducing content
Physical Activity:
- Sharp decline in physical activity
- Rarely leaving the house
- Difficulty in performing basic daily tasks
Productivity:
- Significant decrease in work output
- Struggle to initiate or complete tasks
- Reduced creative output or problem-solving ability
Self-Care:
- Neglect of personal hygiene and appearance
- Irregular eating patterns, often under-eating
Understanding the onset and progression of depressive episodes is crucial for timely intervention. By recognizing the early signs — such as decreased social media activity, withdrawal from communication, and disrupted sleep patterns — I can seek support and implement strategies to manage the episode. While I’m not often depressed, predicting this state was in second place for importance behind mania.
All Together Now
The integration of AI and this sort of data tracking has opened up a new way to understand myself. Comprehensive datasets — our communication, social media activity, health metrics, and more — contain patterns and triggers that precede mood shifts. This data-driven approach offers a more nuanced understanding of the condition and provides actionable insights for managing it — which I’m already putting in place. I found some potential predictive indicators and clear patterns that are making an impact:
Euthymia:
Predictors: Balanced sleep, consistent positive communication tone, moderate social media activity.
Indicators of Shift:
- Increase in Social Media Posts: A marked rise in the frequency of social media posts can indicate an impending shift out of euthymia.
- Frequent Checks of Less-Used Apps: Increased activity on apps like Discord or Telegram, which I usually don’t use as often, serves as a warning sign.
- Rise in Online Purchases: A noticeable increase in online shopping, particularly for non-essential items, signals a potential shift towards hypomania.
- Use of Positive Action-Oriented Language: The language in my communications becomes more positive and action-driven, reflecting heightened mood.
- Variable Location History: More frequent changes in location and visits to new places indicate restlessness, often preceding a mood shift.
- Peaks in Physical Activity: Physical activity, as measured by step counts, tends to peak before falling into periods of hyperfocus, suggesting a transition towards hypomania.
Hypomania:
Predictors:
- Reduced Sleep
- Increased Positive Sentiment: Overall, my sentiment during hypomania is consistently more positive.
- Heightened Communication Across Platforms: There is a marked increase in communication, particularly with an emphasis on professional networks like LinkedIn.
- Erratic Physical Activity: Physical activity levels rise, but in an erratic pattern, reflecting the restless energy characteristic of hypomania.
Indicators of Shift:
- Spikes in Social Media Activity: There is a notable surge in social media engagement, characterized by frequent and enthusiastic posts.
- Assertive and Frequent Messaging: Messaging becomes more frequent, assertive, and sometimes aggressive.
- More New Destinations in Location Tracking: I tend to visit more new places, indicating increased exploratory behavior.
- Periods of No Activity: Twelve-hour periods of no activity, often due to sleep, usually lead to a shift from hypomania to depression, indicating a significant mood transition.
Mania:
Predictors:
- Extreme Lack of Sleep
- Explosive Increase in Digital Footprint: There is a dramatic increase in digital activity across all platforms, including financial transactions and social media posts.
- Erratic Physical Activity: Periods of intense activity followed by prolonged inactivity due to hyperfocus.
Indicators of Shift:
- Peak Communication on Personal Messaging Platforms: Communication frequency reaches its highest on personal messaging platforms, especially social media.
- Increased Purchases: There is a substantial rise in purchases, often impulsive and unnecessary.
- Writing More Code: My GitHub activity spikes as I hyperfocus on new projects.
- Rapid Sentiment Shifts: Sentiment fluctuates rapidly between euphoria and frustration or irritation.
- Minimal Physical Activity: Physical activity drops significantly due to hyperfocus on technology-related tasks.
- Overwhelming Social Media Posting: The volume of posts on social media becomes overwhelming, often incoherent.
- Neglect of Basic Health Needs: Basic needs like food, water, and safety are neglected.
Depression:
Predictors:
- Excessive Sleep
- Significant Drop in Social Media Presence: Social media activity plummets, with minimal engagement.
- Limited Communication: Communication with friends and family becomes sparse, limited to essential interactions.
- Reduced Physical Activity: Physical activity decreases drastically, reflecting the lack of motivation and energy.
Indicators of Shift:
- Withdrawal from Communication: Absence from social interactions, both online and offline.
- Disrupted Sleep Patterns: Sleep patterns become erratic, with periods of excessive sleeping or insomnia.
- Persistent Negative Sentiment: Sentiment remains predominantly negative, characterized by feelings of hopelessness and fatigue.
- Passive Media Consumption: Media consumption becomes passive, with a tendency towards “doomscrolling.”
- Desire for Excessive Sleep: The urge to sleep excessively increases, further indicating the depth of the depressive episode.
General Observations
- My hypomanic and manic states are highly seasonal, clustering around the winter over a long period of time(text message analysis went back twelve years).
- Frequency of communication via messaging and online platforms is my single largest predictor of escalation.
- Mania is a larger financial liability than I realized, and an underappreciated data point.
- Geolocation also proved to be highly informative, which was not my initial assumption.
- Similarly, physical activity escalating through hypomania and then dropping in mania was unexpected, which partially reflects my lack of self awareness in that state.
- Maintaining the online journal was key to the analysis as it provided well anchored timestamps with rich information about my state.
- There’s much more to write, and this approach warrants a more rigorous investigation. I hope others will carry this further while I do myself.
While far from perfect, these insights have helped me to predict multiple potential episodes and shape what they looked like — several times by anticipating forgetting medication, which was a surprising and useful outcome. By continuously monitoring various aspects of my life I’ve gained a much better understanding of myself and my mental health. This continuous feedback loop is the sort of thing I thrive on professionally; I think of it as project managing my own mind.
10/6/2023
Checkpoint in my ongoing bipolar diary — five weeks of consecutive balance! Right around now is when I usually decide I don’t need my medication anymore. So I’m not gonna do that. @coolhand
Of course, it has to be said: The use of AI and extensive health tracking raises significant privacy concerns. The data collected is deeply personal, and its misuse can have serious consequences. It’s also very important to triple check outputs and structure data in specific ways. This is by no means a finished idea, and one that I look forward to evolving. If you’d like to try this yourself, join the Patreon for guidance on extracting your data, prompting an AI, and more experiments like this one. Just make sure you don’t mind having a robot know your step count.
