Our initial exploration of this data shows some interesting trends based on the frequency of the different alarms. Alarms for respective codes make different noises. We can see that some smoothed trend lines are monotonic while others are very non-monotonic. Codes which appear quite non-monotonic had a much lower frequency than other codes. There also appear to be some plateaus and regressions in the learning curves as time progress. It appears there might be a relationship between the total number of a certain codes played and a slower response time showing that participants are becoming conditioned to certain tones played by the XVAD device.  

Negative Relationship between Alarm Response Time and Frequency. As the number of alarms responded to increases, there is a corresponding increase in response time, indicating a negative correlation. The intercept is estimated to be 5471.51 milliseconds and the slope coefficient for the predictor variable (response) is estimated to be -93.67 milliseconds. Both the intercept and the slope coefficient have p-values less than 0.05, indicating that they are statistically significant. The R-squared value is 0.1302, indicating that about 13% of the variability in the response variable can be explained by the predictor variable. The F-statistic is 15.26 with a p-value of 0.0001683, indicating that the model is statistically significant and that it provides a good fit to the data. Based on these results, we can conclude that there is a statistically significant negative linear relationship between the predictor variable x and the response variable y. For every one unit increase in an alarm, we expect to see a decrease of approximately 93.67 milliseconds. 

This is a logistic regression model with the response variable Pass and fail and predictor variable Time, using a binomial family. The model shows that the predictor variable Timelog has a positive coefficient of 3.7973, which means that for each one-unit increase in Timelog, the log-odds of a failing response increase by 3.7973. Both coefficients are statistically significant (p < 0.05), indicating that the relationship between the predictor variable and response variable is unlikely due to chance. Based on the findings from this logistic regression model, we can conclude that the log of the Time variable is a significant predictor of whether a participant will pass or fail to respond to an alarm. Additionally, the intercept of -33.7316 suggests that the log-odds of failing is very low when Timelog is zero, which may indicate a baseline level behavior before any intervention or treatment.

Bootstrapping is a powerful resampling technique that has gained popularity due to its ability to provide reliable estimates of standard errors and confidence intervals for various statistics, particularly in situations where direct standard error formulas are complex or unavailable. The fundamental concept of bootstrapping revolves around the notion that drawing conclusions about a population using sample data can be achieved by resampling the sample data and conducting inferences about a sample derived from the resampled data (resampled → sample → population).  We see the data starts form into a 'normal distribution' shape as we add more data points.

The results show an almost normal distrubiton. The black line is the boot strapped data and the red line is a normal distribution. It's clear they are very similar. Normal distributions are commonly observed in many natural and social phenomena, such as heights and weights of people, exam scores, and many physical and biological measurements. They are important because they have several properties that make them easy to work with mathematically and statistically. For example, the mean, median, and mode of a normal distribution are all equal, and the proportion of the data within a certain number of standard deviations of the mean can be easily calculated using the empirical rule or z-scores. 

When a dataset takes a normal distribution shape, it is often possible to use statistical tests and models that assume a normal distribution, such as the t-test, ANOVA, and linear regression. However, it is important to note that not all datasets follow a normal distribution, and it is important to check for normality and explore other distributional assumptions before applying these methods.

Confidence Interval results:

   5%        50%        95%

4676.242 5116.081   5525.220 

So, we can be 95% confident that, on average, XVAD users will respond to the XVAD alarms between 4676 and 5525 milliseconds. This is great as it means most users will respond at a passing rate. 

Following phase 1, some participants engaged in focus groups. Sessions were recorded with consent and sentiment analysis of the transcribed discussions was performed. The sentiment analysis was performed on the transcribed discussions of focus groups to gain deeper insights into the participants' experience. Overall, the sentiment scores were positive across all groups. Sentiment analysis is a process of extracting subjective information from text data. It is a computational approach used to determine the attitude or opinion expressed in a piece of text, whether it is positive, negative, or neutral. Because the sentiment scores were heavier on the positive scale across all groups, we can infer that the participants had a positive attitude towards the topic of discussion. This is a good indication that the processes implemented has worked well. 

The sentiment analysis process involves several steps. First, the text data is preprocessed, which involves removing stop words, punctuation, and other irrelevant information. Next, the text is tokenized, which involves splitting the text into individual words or phrases. Then, the sentiment of each token is determined based on the context in which it appears, using methods such as rule-based systems, machine learning models, or lexicon-based approaches.  Once the sentiment of each token has been determined, an overall sentiment score can be calculated for the entire text. This can be done by summing the sentiment scores of all tokens. 

For sentiment analysis to work well, stop words such as; like, the, and, of, etc. should be removed. Additionally, domain specific words like; 'alarm' should be removed as it likely refers to the task reminder, not causing the sensation of being alarmed or something being alarming. Removing domain-specific and stop words is beneficial for it improves the accuracy of models that process text data by removing irrelevant terms that may be specific to a particular field or industry and it can make the text more easily understandable to people who are not familiar with the specific field or industry being analyzed.