Report Working From Home

General Descriptives

need for correlation tables.

Descriptives over Time

Below you find the reported burnout index for home (november) versus work (pre-corona). Standard deviations are applied.

Time series for the major outcome variables

Timeseries of self-reported Single Questions

In the figure below we see the path over time for self-reported, single question metrics about the satisfaction of our participants to work from home. These metrics include: Satisfaction to work from home (how satisfied are you) and Happiness to work from home (how happy are you). For the June measurement, we see that happiness to work from home steeply declined during the lockdown, but left in June at a reasonable level. The satisfaction overall only rose. Interestingly, in november, all participants expected the lockdown period and june to be worse than it was than (retrospectively pessimistisch), and never returned to the satisfaction level of june. No difference in trend between june and november. Thus, in november, people report an increase in satisfaction and happiness comparable to june, but throughout the board, are less happy and satisfied compared to june (note that the graph displays error bars, not sd)

Regressions

November

First, we look at the simple linear regressions on November factors on Productivity and Burn-out inventory. Center of Building Environment (CBE) variables include the Temperature, Air Quality, Light, and Noise Satisfaction. Home-working condition (HWC) variables pertain to the satisfaction of home office factors: Desk, Chair, Screen, Hardware, and Wifi.

We add a multitude of personal, family, and office conditions factors to control for in our analysis. On a personal level, we include gender, age, education level. For family, we add the family size and dummies for children without care during working hours, and children with care during working hours, with a baseline of no children. For office conditions, we finally include dummies for bad and good natural light, both for which neutral lighting is included in the baseline. We also include ventilation time in %, which indicates the percentage of total time at work in which the (home) office is ventilated (either mechanically or manually). Finally, we control for the self-reported degree of which the participant states that their job is suited to work from home.

Figure 1.a shows that for both temperature and noise satisfaction, higher scores significantly increase productivity and decrease burnout index. For the other CBE variables, we find that air quality satisfaction lowers burnout inventory, and light satisfaction only increases productivity.

Figure 1.b shows that HWC variable satisfaction has less influence on productivity and burnout: desk satisfaction increases productivity, as does hardware satisfaction. Only Wifi satisfaction both increases productivity whilst lowering burnout index. Both screen and chair satisfaction are not related to our outcome metrics.

As a side note, we see in the controls of both our models (2.c and 2.d) that having to take care of children during working hours has a profound negative effect on both metrics (increasing burnout, and decreasing productivity), more than just having kids without care.

In the following regression figure 3 we add the factor of Stress and Irritability. We see that Irritability has a high correlation with Burnout regarding the effect on our dependent variables.

For completion, figure 4 below shows all 5 clusters in november on CBE and HWC factors.

June to November

In the graph below, we compare the regression coefficients as they move from June to November. Overall, we observe stronger signs for each of the independent variables. Compared to June, November scores are more strongly predicting productivity. Thus, in November, the satisfaction with CBE and HWC factors becomes more important / stronger related to productivity.

Note that we cannot create such a plot for the burnout metric, as we did not measure burnout during june. We have only asked our participants to state their current burnout levels, and their burnout levels pre corona.

Precorona basic situation.

In this model, we assess the base situation of all participants. We ask the CBE and HWC factors (note: the conditions at their office)

We suffer from some annotation problems. Here follow the seperate figures of 6:

Precorona to Working from Home (Delta’s)

Here we finally look at the delta’s (differences) of CBE and HWC factors between home office and work office. We test whether the increased or decreased satisfaction on CBE and HWC (controlled for the base level at work) explains the satisfaction level at the home office.

The effect of ventilation group on outcome metrics

In this part, we split our sample according to the distribution of the ventilation rate during the day. Note that we ask participants to rate the percentage of time they ventilate (either mechanical or physical) during the whole day. 100% equals the full day, 0% means not at all during the day.

Descriptives by Ventilation

Below you find the total and the

Summary Statistics of WfH Per Ventilation Group

	Low Ventilation (N=320)	Medium Ventilation (N=346)	High Ventilation (N=336)	Total (N=1002)	p value
Percentage of Ventilation per day					< 0.001 (1)
- Mean (SD)	5.853 (4.167)	34.725 (12.805)	89.851 (12.974)	43.990 (36.318)
- Median (Q1, Q3)	5.000 (1.000, 10.000)	30.000 (25.000, 50.000)	100.000 (80.000, 100.000)	30.000 (10.000, 80.000)
- Min - Max	0.000 - 10.000	15.000 - 50.000	60.000 - 100.000	0.000 - 100.000
Job fit for WfH					0.173 (1)
- Mean (SD)	7.616 (2.213)	7.494 (2.419)	7.679 (2.514)	7.595 (2.387)
- Median (Q1, Q3)	8.000 (7.000, 9.000)	8.000 (7.000, 9.000)	8.000 (7.000, 10.000)	8.000 (7.000, 10.000)
- Min - Max	1.000 - 10.000	1.000 - 10.000	1.000 - 10.000	1.000 - 10.000
Kids at Home					0.159 (2)
- Always	15 (4.7%)	10 (2.9%)	8 (2.4%)	33 (3.3%)
- Sometimes	95 (29.7%)	128 (37.0%)	110 (32.7%)	333 (33.2%)
- Never	43 (13.4%)	52 (15.0%)	59 (17.6%)	154 (15.4%)
- No Kids	167 (52.2%)	156 (45.1%)	159 (47.3%)	482 (48.1%)
- N-Miss	0	0	0	0
Willingness to continue WfH					< 0.001 (1)
- Mean (SD)	5.797 (2.949)	6.136 (2.930)	6.792 (2.843)	6.248 (2.933)
- Median (Q1, Q3)	6.500 (3.000, 8.000)	7.000 (3.000, 8.000)	7.000 (5.000, 9.000)	7.000 (4.000, 9.000)
- Min - Max	1.000 - 10.000	1.000 - 10.000	1.000 - 10.000	1.000 - 10.000
WfH experience					0.245 (2)
- No	89 (46.1%)	95 (43.8%)	113 (51.6%)	297 (47.2%)
- Yes	104 (53.9%)	122 (56.2%)	106 (48.4%)	332 (52.8%)
- N-Miss	127	129	117	373
Gender					0.097 (2)
- Male	171 (53.4%)	204 (59.0%)	207 (61.6%)	582 (58.1%)
- Female	149 (46.6%)	142 (41.0%)	129 (38.4%)	420 (41.9%)
- N-Miss	0	0	0	0
opleiding					0.083 (1)
- Mean (SD)	7.416 (2.859)	7.101 (2.870)	6.914 (2.816)	7.139 (2.853)
- Median (Q1, Q3)	7.000 (5.000, 11.000)	7.000 (4.000, 11.000)	7.000 (4.000, 10.000)	7.000 (4.000, 11.000)
- Min - Max	2.000 - 11.000	2.000 - 11.000	2.000 - 11.000	2.000 - 11.000
Age					0.351 (1)
- Mean (SD)	43.144 (12.228)	44.029 (13.235)	44.464 (12.087)	43.892 (12.539)
- Median (Q1, Q3)	44.000 (32.000, 53.000)	44.000 (32.000, 56.000)	45.000 (34.000, 54.000)	44.000 (33.000, 54.000)
- Min - Max	20.000 - 67.000	19.000 - 67.000	20.000 - 68.000	19.000 - 68.000

1. Kruskal-Wallis rank sum test
2. Pearson’s Chi-squared test

Plots on CBE per group

First we look at the spider graph for the difference at home. Note that Outer is better ventilated, inner is worst venitlated (legend fix underway)

Next, we show the lollipop graphs per ventilation split. Note here that the results are naturally ordered (so not similar order in each table). We generally observe a improvement of CBE satisfaction per ventilation group (note: no statistical test yet). Most importantly, we see that there is no difference in office metrics. This alleviates the worry that the high ventilation group simply answers high on default (high ventilation, high on satisfaction).

Warning: `add_rownames()` was deprecated in dplyr 1.0.0.
Please use `tibble::rownames_to_column()` instead.

Statistical Analysis for Mean differences CBE Satisfaction

Satisfaction in CBE

Here we see the statistical proof of the difference between CBE satisfaction at work and in the homeoffice. Below, you find the results of a MANOVA, and means plots. For the CBE at home, we see that ventilation makes a difference. Specifically, looking at the TUKEY post-hoc results, we find that all differences are statistically significant.

fullsat <- read.csv("~/Dropbox/Project 4 - COVID/Data/Homework/fullsat.csv", sep=",", stringsAsFactors=TRUE)
fullsat$x <- as.factor(fullsat$x)

## good sample set __> MANOVA LETS GO


CBE_Man <- manova(cbind(CBE_H1, CBE_H2, CBE_H3, CBE_H4) ~ x, data = fullsat)

summary.aov(CBE_Man)

 Response CBE_H1 :
             Df  Sum Sq Mean Sq F value    Pr(>F)    
x             2   44.61 22.3041  14.027 9.823e-07 ***
Residuals   999 1588.53  1.5901                      
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

 Response CBE_H2 :
             Df  Sum Sq Mean Sq F value    Pr(>F)    
x             2   76.42  38.211  32.305 2.541e-14 ***
Residuals   999 1181.63   1.183                      
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

 Response CBE_H3 :
             Df  Sum Sq Mean Sq F value    Pr(>F)    
x             2   67.87  33.936  24.478 4.183e-11 ***
Residuals   999 1384.97   1.386                      
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

 Response CBE_H4 :
             Df  Sum Sq Mean Sq F value    Pr(>F)    
x             2   48.55 24.2727   14.29 7.608e-07 ***
Residuals   999 1696.95  1.6986                      
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

438 observations deleted due to missingness

CBE11 <- plotmeans(fullsat$CBE_H1 ~ fullsat$x)

CBE12 <- plotmeans(fullsat$CBE_H2 ~ fullsat$x)

CBE13 <-plotmeans(fullsat$CBE_H3 ~ fullsat$x)

CBE14 <-plotmeans(fullsat$CBE_H4 ~ fullsat$x)

aov(CBE_H1 ~ x, data = fullsat)

Call:
   aov(formula = CBE_H1 ~ x, data = fullsat)

Terms:
                        x Residuals
Sum of Squares    44.6082 1588.5256
Deg. of Freedom         2       999

Residual standard error: 1.260998
Estimated effects may be unbalanced
438 observations deleted due to missingness

CBE_Man1 <- aov(CBE_H1 ~ x, data = fullsat)
CBE_Man_Tuk <- TukeyHSD(CBE_Man1)
CBE_Man_Tuk

  Tukey multiple comparisons of means
    95% family-wise confidence level

Fit: aov(formula = CBE_H1 ~ x, data = fullsat)

$x
         diff        lwr       upr     p adj
2-1 0.2764632 0.04690667 0.5060196 0.0132924
3-1 0.5215774 0.29038540 0.7527694 0.0000004
3-2 0.2451142 0.01841504 0.4718134 0.0303604

Hinder in CBE

We indeed see that the CBE conditions at work are insignificant.

CBE_ManW <- manova(cbind(CBE_W1, CBE_W2, CBE_W3, CBE_W4) ~ x, data = fullsat)
summary(CBE_ManW, intercept=TRUE)

             Df  Pillai approx F num Df den Df Pr(>F)    
(Intercept)   1 0.95658   5485.0      4    996 <2e-16 ***
x             2 0.00307      0.4      8   1994 0.9299    
Residuals   999                                          
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

# Univariate models 
summary.aov(CBE_ManW)

 Response CBE_W1 :
             Df  Sum Sq Mean Sq F value Pr(>F)
x             2    2.15  1.0745  0.6971 0.4983
Residuals   999 1539.90  1.5414               

 Response CBE_W2 :
             Df  Sum Sq Mean Sq F value Pr(>F)
x             2    1.94 0.96822  0.6001  0.549
Residuals   999 1611.82 1.61343               

 Response CBE_W3 :
             Df  Sum Sq Mean Sq F value Pr(>F)
x             2    1.01 0.50562  0.3009 0.7402
Residuals   999 1678.67 1.68035               

 Response CBE_W4 :
             Df  Sum Sq Mean Sq F value Pr(>F)
x             2    3.58  1.7920  0.9944 0.3703
Residuals   999 1800.31  1.8021               

438 observations deleted due to missingness

CBEW11 <- plotmeans(fullsat$CBE_W1 ~ fullsat$x)

CBEW12 <- plotmeans(fullsat$CBE_W2 ~ fullsat$x)

CBEW13 <- plotmeans(fullsat$CBE_W3 ~ fullsat$x)

CBEW14 <- plotmeans(fullsat$CBE_W4 ~ fullsat$x)

Same Plot as above for Hinder

Note that

Below you find the difference score for the CBE metrics. Note that these results are split for ventilation groups. For each ventilation group, we look at the CBE factor, and see the improvement or deterioration moving from the office to home. Overall, we see that all groups prefer working from home more, and better ventilated groups find this improvement better.

Same plots now for HWC

Here we show the effect of ventilation on HWC. By looking at the effect of ventilation on HWC, and comparing these effects with CBE, we are able to say something about the overall effect of ventilation on behavior. If ventilation only has an effect on CBE, ventilation hints to only effects the overall IEQ. In this case, ventilation proxies the overall satisfaction with all objective IEQ such as noise, air quality, and temperature (also light). When ventilation has the same or similar effect on HWC, we can safely assume that ventilation improves a more general satisfaction. HWC are generally independent of ventilation rate, yet an improvement on these metrics for

HWC_Man <- manova(cbind(HWC_H1, HWC_H2, HWC_H3, HWC_H4) ~ x, data = fullsat)
summary(HWC_Man, intercept=TRUE)

             Df  Pillai approx F num Df den Df    Pr(>F)    
(Intercept)   1 0.94630   4388.1      4    996 < 2.2e-16 ***
x             2 0.06997      9.0      8   1994 2.942e-12 ***
Residuals   999                                             
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

# Univariate models 
summary.aov(HWC_Man)

 Response HWC_H1 :
             Df  Sum Sq Mean Sq F value    Pr(>F)    
x             2   88.48  44.239  19.174 6.744e-09 ***
Residuals   999 2305.01   2.307                      
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

 Response HWC_H2 :
             Df  Sum Sq Mean Sq F value    Pr(>F)    
x             2  105.45  52.725  21.919 4.827e-10 ***
Residuals   999 2403.04   2.405                      
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

 Response HWC_H3 :
             Df  Sum Sq Mean Sq F value    Pr(>F)    
x             2   62.68 31.3394  13.765 1.267e-06 ***
Residuals   999 2274.48  2.2768                      
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

 Response HWC_H4 :
             Df  Sum Sq Mean Sq F value    Pr(>F)    
x             2   83.65  41.825  25.004 2.535e-11 ***
Residuals   999 1671.08   1.673                      
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

438 observations deleted due to missingness

HWCH11 <- plotmeans(fullsat$HWC_H1 ~ fullsat$x)

HWCH12 <- plotmeans(fullsat$HWC_H2 ~ fullsat$x)

HWCH13 <- plotmeans(fullsat$HWC_H3 ~ fullsat$x)

HWCH14 <- plotmeans(fullsat$HWC_H4 ~ fullsat$x)

HWC_ManW <- manova(cbind(HWC_W1, HWC_W2, HWC_W3, HWC_W4) ~ x, data = fullsat)
summary(HWC_ManW, intercept=TRUE)

             Df  Pillai approx F num Df den Df  Pr(>F)    
(Intercept)   1 0.97173   8560.5      4    996 < 2e-16 ***
x             2 0.01519      1.9      8   1994 0.05483 .  
Residuals   999                                           
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

# Univariate models 
summary.aov(HWC_ManW)

 Response HWC_W1 :
             Df  Sum Sq Mean Sq F value   Pr(>F)   
x             2   14.11  7.0569  5.7265 0.003366 **
Residuals   999 1231.09  1.2323                    
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

 Response HWC_W2 :
             Df  Sum Sq Mean Sq F value  Pr(>F)  
x             2    8.38  4.1923  3.2681 0.03849 *
Residuals   999 1281.51  1.2828                  
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

 Response HWC_W3 :
             Df  Sum Sq Mean Sq F value  Pr(>F)  
x             2   10.57  5.2828  4.4368 0.01207 *
Residuals   999 1189.49  1.1907                  
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

 Response HWC_W4 :
             Df  Sum Sq Mean Sq F value   Pr(>F)   
x             2   11.65  5.8233  5.1453 0.005982 **
Residuals   999 1130.64  1.1318                    
---
Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1

438 observations deleted due to missingness

HWCW11 <- plotmeans(fullsat$HWC_W1 ~ fullsat$x)

HWCW12 <- plotmeans(fullsat$HWC_W2 ~ fullsat$x)

HWCW13 <- plotmeans(fullsat$HWC_W3 ~ fullsat$x)

HWCW14 <- plotmeans(fullsat$HWC_W4 ~ fullsat$x)

Plot On performance metrics

Note that the first plot does not include burnout metrics, the bottom one does include Burnout.