1. Effect of IEQ on work productivity
Alan Hedge & Daniel Gaygen
Cornell University
and James Smith
Syracuse University 1

2. Productivity Measures
Study Goals (‘03)
Collect data on IEQ variables and work productivity for specific workstations every 15 minutes for a minimum of 1 week per worker.
Model development - model how work productivity can be predicted as a function of pollutant levels in the workstation microenvironment.
Estimate the economic impacts of IEQ variations on work productivity.
IEQ Measures
TVOC
PM10
CO2 Ta RH
Productivity Measures
# correct keystrokes/minute
# error keystrokes/minute
Mouse use 2

3. Examples of IEQ Unit placement

4. Research Methods
Synchronous data on work performance and indoor environment will be collected at workstations for at least 1 work week (now considering 2 weeks minimum) per worker for a total of 100 workers.
IEQ (TVOC, PM10, T, RH, CO2,) will be monitored using portable data logging systems.
Subjects will be workers working on computer systems where their employer is using software to measure their work performance (# of correct keystrokes, # error keystrokes per minute, and time spent using the mouse).
Groups of 12 workstations (12 sets of IAQ equipment are available) will be tested at a time (subjects will be working on computers, doing similar office work and in cubicle office space).
Privacy and confidentiality of worker information will be preserved at all times.
In each building, a series of locations that reflect a variety of environmental conditions will be identified before deploying the data loggers. 4

5. Research Challenges
Equipment problems: reliably functioning units not available until Winter Currently 1 unit remains out of operation since the start of 2007 (11 still functional).
All previous field site negotiations ultimately proved unfruitful.
Company marketing the software that field site companies were using or planning to use underwent corporate buyout and transformation.
Alternative plan formulated and implemented in Spring ‘07.

6. Office Environment Study
Law office – suite in office building
Garden City, NY
Opposite construction site and busy highway
Office workers complaining of thermal discomfort issues
Office survey in March/April ‘07

7. Office Environment Study
Free-standing furniture office setup
16 employees (1 left half way through study)
14 women, 2 men (complete data for 13 women, 2 men)
6 lawyers, 10 administrative support
Every computing minute (keying/mousing) logged via web-based system
8am to 6pm data analyzed
90,467 data points

8. IEQ Survey Plan 6 had IEQ units for 4 weeks
5 had IEQ units for 2 weeks and then HOBO H8 for 2 weeks
5 had HOBO units for 2 weeks and then IEQ units for 2 weeks
Weeks 1 & 2
Weeks 3 & 4

9. Environmental Conditions: Temperature

10. Effect of Temperature on Performance [1]

11. Effect of Temperature on Performance [2]

12. Environmental Conditions: Humidity

13. Effect of RH on Performance [1]

14. Effect of RH on Performance [2]

15. Environmental Conditions: CO2 Concentration

16. Effect of CO2 on Performance [1]

17. Effect of CO2 on Performance [2]

18. Environmental Conditions: PM10

19. Effect of PM10 on Performance [1]

20. Effect of PM10 on Performance [2]

21. Environmental Conditions: TVOC

22. Effect of TVOC on Performance [1]

23. Effect of TVOC on Performance [2]

24. Mixed Models Analysis
Mixed models are a recent extension of the General Linear Model
Mixed models are also called hierarchical linear models (HLM)
Mixed Models contain both fixed and random effects
Fixed Effects: factors for which the only levels under consideration are contained in the coding of those effects (e.g. gender)
Random Effects: Factors for which the levels contained in the coding of those factors are a random sample of the total number of levels in the population for that factor (e.g. subject)
Mixed models are ideal for with data with non-constant variability
All statistical analyses undertaken with assistance from Françoise Vermeylen of the Cornell Statistical Consulting unit.
All initial models included:
Fixed factors: gender, job, weekday, time-of-day, and all environmental variables
Random factors: subject, studyday

25. Distribution of Log-Mean Correct Keystrokes

26. Distribution of Log-Mean Correction Keystrokes

27. Distribution of Mean Left-Button Mouse Clicks

28. Results
Contemporaneous Environmental Effects on Correct Keying
Significant effect of contemporaneous air temperature on the ln hourly mean correct keystrokes/minute (F[1, ] = 4.689, p= 0.031)
Hourly mean correct keystrokes/minute = e[ °C].
No other environmental, temporal or demographic variables were significant. The analyses were run for the environmental data lagged at 1 and 2 hours and no significant effects were found.
ln hourly mean correct keystrokes /minute

29. Results
Contemporaneous Environmental Effects on Mousing
Significant effect of contemporaneous CO2 concentration (ppm) on the hourly mean left-button single mouse clicks/minute (F[1, ] = , p= 0.001)
Hourly mean left-button single mouse clicks/minute = e[ ppm].
No other environmental, temporal or demographic variables were significant. The analyses were run for data lagged at 1 and 2 hours and no significant effects were found.
ln hourly mean left mouseclicks/minute

30. Results
Contemporaneous Environmental Effects on Correction Keystrokes
There were no contemporaneous effects of any environmental variable on log-mean correction keystrokes/minute.

31. Lagged Environmental Effects on Correction Keying
For the 1 hour lagged data there was a significant effect of hourly mean CO2 (F[1, = 4.807, p = 0.029) and a significant linear effect of mean PM10 (F[1, = 6.102, p = 0.014) on ln mean correction keystrokes/minute.
The main effect of CO2 was curvilinear and subsequently the data were partitioned at 425ppm CO2 and separate mixed model analyses were run.
There were 119instances where the mean hourly CO2 level was <= 425 ppm and there was no significant effect of either CO2 or PM10
There were 993 instances when the CO2 level was greater than 425 ppm was a significant linear fit for 1-hour lagged CO2 (F1, = 6.151, p = 0.014) and 1-hour lagged PM10 (F1, = 6.418, p = 0.012) and the resulting equation is:
Hourly mean correction keystrokes/minute = e[0.0039(lag1CO2) (lag1PM10)]

32. Lagged PM10 Effect on Correction Keying
ln hourly mean correction keystrokes /minute

33. Lagged CO2 Effects on Correction Keying
ln hourly mean correction keystrokes/minute

34. Conclusions
Associations between the quantity and quality of computer work and indoor environmental conditions can be measured.
Work performance costs of poor indoor environmental conditions can be measured.
Results support the concept of an optimal zone for IEQ.

35. Immediate Future Plans
Advanced negotiations with a major corporate offices in Piscataway, NJ; Syracuse and Manhattan for next field test sites.
Looking for sites with IAQ issues.
Possibility of interventions in existing garden City site.
Performance monitoring deployable to any internet-linked organization.
The number of IEQ monitoring units is a bottleneck.

36. VOC Effects on Higher Cognitive Processes
SAC previously expressed concerns about also studying the effects of IAQ on
higher cognitive processes.
Early in 2006 we had equipment issues and no imminent test sites, so we ran a laboratory study of the effects of a VOC mixture (complex aromatic mixture of natural plant extracts with over 200 organic compounds) on higher cognitive task performance (lexical access).
Lexical decision task is a standard paradigm for investigating lexical access, which occurs when humans recognize a word and its meaning and syntactical properties are available for use.
Lexical access was inhibited by VOC exposure and participants responded more slowly and less accurately to words, especially to low frequency words.
However, the results were asymmetric and the effect was only statistically significant with VOC exposure in the second session after no VOC in the first session. Encoding specificity may explain the results.
Poster provides more information.