Participatory Governance and Responsiveness: Do Motivational Interventions Increase Engagement with Citizen Input?

Public Sector Values

A lot of work on bureaucracy has focused on personnel selection. Choosing the people with the right set of skills and motivations should lead to more efficient and effective delivery of public services [e.g. see Bertelli (2012) for a more in-depth discussion]. However, bureaucrats’ performance can change with age and tenure, even when there is no reform to public management (Esteve and Schuster, 2019). Thus, not personal characteristics but sets of values and motivations might be a better predictor of engagement. It has been posited that bureaucrats might possess a certain type of motivation that predicts performance in the absence of monetary rewards: public service motivation.

Public service motivation is a set of values that encourages expending effort to provide high quality services to the public, usually in absence or unresponsive to monetary incentives (Perry, 2015). In its original conceptualisation, it encompasses (i) attraction to public policy making, (ii) commitment to the public interest, (iii) civic duty, (iv) social justice, (v) self-sacrifice and (vi) compassion. Particularly, (ii) commitment to the public interest should motivate bureaucrats to provide access responsiveness to citizens.

In economics, values have been recognised as factors key to understanding the behaviour of individuals and groups. Values shape other– and process–regarding behaviours alongside more classically rational, purely selfish behaviours (Ben-Ner and Putterman, 1998). Research on public service motivation suggests that bureaucrats might be particularly inclined to act in the interest of the public because of certain values they hold (Meier and O’Toole, 2006; Perry, 2015; Ritz, 2015; Tummers and Bekkers, 2014). Research from organisational and clinical psychology, which suggest that seeing an explicit link between one’s values and one’s work can improve motivation and performance – a finding that holds true across industries and geographical contexts (Maricu¸toiu et al., 2016). It appears plausible that highlighting the link between engaging with citizen input and values that bureaucrats hold could increase their motivation.

Yang and Callahan (2007), find that value congruence is important for bureaucrats to support citizen initiatives. In a survey of U.S. public administrators, they find that bureaucrats respond more to participatory input when they value citizen input. Further, departments that have a citizen-or client-orientation are also more receptive to input. It is plausible that when citizen input highlights key values of bureaucrats they are more likely to engage. However, since studies such as Yang and Pandey (2011) are forced to rely on self-reports, it is unascertained whether observed links are due to social desirability and whether such correlations would hold if we looked at actual use (a behavioural measure) instead of self-reported behaviour. Further, it remains unclear whether a causal relationship exists: do the values bureaucrats hold lead them to be more open to considering citizen input when it reflects their values?

As Bakker (2015) elaborates, public sector motivation and its antecedents can buffer against organisational stressors and facilitate engagement.

In line with evidence on value-congruence, it is expected that:

Input stressing core values of public service – caring for the opinions of the citizens one serves – will lead to (H1) greater responsiveness than messages lacking such qualities.

Symbolic Rewards

The motivating effect of public service values might be limited to certain subsets of bureaucrats (Linos, 2018; Asseburg et al., 2018). Bureaucrats, like private sector workers, also respond to extrinsic rewards such as pay for performance (Andrews and Esteve, 2015; Esteve and Schuster, 2019). However, in the public sector, the scope for monetary incentives is limited. Non-monetary rewards, typically called ‘symbolic rewards’ in the economics literature, are more easily available. They are hypothesised to affect motivation through social recognition of performance, both through activating immediate gratification (feeling rewarded) and longer-term prospects (a kind of halo effect that induces others to award more attention and praise to the recipient of an award; see Frey and Gallus (2017) for a more extensive discussion).

Symbolic rewards have been employed across settings and seen promising results. For instance Panagopoulos (2011) found that one of the simplest forms of a symbolic reward, a ‘thank you’ message, could increase turnout in three separate U.S. elections. Such behaviour does not appear to be limited to the electoral context. It has also been observed in a variety of organisational settings (Grant et al., 2007; Bradler and Neckermann, 2016; Gallus, 2015; Bell´e, 2015; Bovaird and L¨offler, 2009; Ashraf et al., 2014). ²

Based on promising results from experiments using symbolic rewards, it is expected that

Input containing an explicit “thank you” (symbolic reward), will lead to (H2a) greater responsiveness than the control email and (H2b) the short- ened and simplified input.

Cognitive Capacity

Motivation can however only affect performance if there is enough cognitive capacity left to deploy. Even the most committed and driven bureaucrat is limited by their cognitive abilities. In the context of budget cuts to government services, fewer staff have to deal with a greater number and variety of requests. If bureaucrats do not have the time and cognitive resources to read through a lot of information, the effect of improving motivation will be limited at best. Instead, reducing the time and cognitive resources it takes to process information could increase the rate at which bureaucrats engage with it. Cognitive load theory predicts that if information load exceeds the available bandwidth, motivation no longer has an effect as there are simply too few cognitive resources available to perform the task or improve performance. On the other hand, when information load is low more bandwidth is available to perform tasks (Moreno and Park, 2010, p.10). Based on this, it could be expected that if citizen input is condensed, so that reading it is quicker and cognitively less burdensome, bureaucrats should be more likely to engage with the information.

(H3a) ³ Reducing cognitive load by shortening a message will increase engagement. This is hypothesised to take place as it increases the likelihood of bureaucrats getting to the end of the message and having time to engage beyond reading the message by clicking on the link and reading on.

However, reducing cognitive load is likely insufficient to produce meaningful and sustained engagement. As Bakker (2015) illustrates, recognising performance and nurturing public service values can be theorised as a central reinforcement mechanism to sustain motivation and performance. (H3b) In line with such reasoning, I expect that motivational messages (value-congruent framing and symbolic rewards) will lead to greater engagement compared to the effects of purely reducing cognitive load.

The next section describes a randomised field experiment used to test these hypotheses.

Study Setting

The study took place in fall 2018. It used a pre-registered randomised field experiment to test the effect of different ways of presenting input derived from citizen participation on access responsiveness. ⁴ The participating local authority (LA) has been a strong-hold of the UK’s Conservative Party since the Second World War. Its population is ethnically more homogeneous and affluent than the England average. The LA comprised several organisations and a wide range of services, which is typical for large UK local authorities. The LA had undergone a change in leadership in the spring of the same year and – like all other local authorities in England and Wales – was focused on re-designing services to cope with pressures brought on by substantial cuts in central government funding. Over the summer of 2018 the LA engaged in a large-scale effort to collect input from citizens resident in its jurisdiction, to develop a strategy to renew and deliver services over the next decade. The engagement exercise comprised of online surveys, town hall meetings, face-to- face interviews and focus group discussions. The authority’s outreach team collected the information in conjunction with a marketing agency which had been hired for the purpose. The internal outreach team condensed the information, whilst the LA’s communications team edited the report and compiled messages for internal communication of the results. The final product was planned to be disseminated via emails to all relevant staff. The purpose was to inform staff about what citizens saw as priorities for the change strategy of the LA and encourage them to work towards delivering on these preferences in their respective areas of work.

I exploited this as an opportunity for a randomised field experiment. ⁵ A few months before the citizen engagement exercise took place, an internal staff consultation on the current state of affairs and pending changes had concluded. Staff could feed back via an anonymous online survey which consisted of open-ended questions and several multiple choice questions, asking staff to rate different aspects of the local authority as a work place. The local authority shared these results with me. The results revealed that staff was highly cynical about previous and impending organisational change, frustrated with an — allegedly — insular style of working, high demands and low reward offered by their jobs. Barriers for engagement with citizen input, which directly related to new organisational changes, were thus notably high. The LA had planned to inform all of its staff of the results of the citizen engagement exercise via an organisation-wide email campaign. The date for dissemination was chosen in a manner that council leaders had the chance to view draft summaries first, but the information had not been spread beyond the outreach team and a few members of the executive. This meant that the information was new to almost all of the bureaucrats employed by the LA and there was scope for them to engage and react to this information. On the web page where information was summarised, actionable points were summarised for each area of work (e.g. education, policing, traffic) to make it evident to staff what was immediately relevant to their role.

Interventions

All interventions were based on the literature reviewed above. The shortened treatment reduced the word count of the control email by 58% and simplified its language. All other treatment emails used the same text as the shortened treatment apart from the email subject line and the first sentence, which contained the treatment message (see Table 1 for treatment messages and subject lines; please refer to the online appendix for the full treatment materials).

OPEN IN VIEWER

Table 1.

Overview of treatment texts.

Exp. Condition	Subject Line	Key message	Word Count
Control	Results of resident and partner engagement on [local authority's] Vision [future year]	N/A	391
Shortened	Results of resident and partner engagement Vision [future year]	N/A	162
Value	Views from the people you care about – [local authority] Vision [future year]	We know you care deeply about the people you serve through your work every day. We want to share with you what they value about [local authority] and what they would like to change.	176
Symbolic	Thank you for your effort in making [local authority]  a great place – [local authority]  Vision [future year]	We would like to thank you for the effort you have put in to make [local authority] a great place to live. We want to share with you what they value about [local authority] and what they would like to change.	178

Implementation

I compare all interventions against how the LA would have disseminated the information without intervention. The status-quo way of disseminating citizen input in the local authority, in which the experiment takes place, is typical for the UK and most authorities in developed countries. Typically, input is collected and condensed by a central strategy or citizen outreach team. This output is reviewed and signed off by senior bureaucratic and political leadership. It then is sent to relevant bureaucrats via email, often including links to more detailed information.

With the help of an email management system and bespoke website analytics, I tracked the behaviour of bureaucrats in response to such emails. All emails contain a summary of the main findings and a link to an intranet page that contains more detail on the findings, broken down by area of work to highlight relevance (e.g. Children’s Social Care, Transport and Planning). The summary highlights key findings from a citizen engagement exercise that the local authority ran in the summer leading up to the trial. Between 3000 and 4000 citizens fed in via online surveys, street interviews, town hall debates and meetings co-organised with community groups. The engagement exercise focused on citizen’s needs, wishes and ideas for an overhaul of the local government’s strategy that was to be implemented within the next 10 years.

Dependent Variables

Bureaucrats reading and engaging with information from citizens is the necessary first step to be responsive to them. Schneider (1999) calls this ‘access responsiveness’: the first necessary condition for there to be responsiveness is for citizens concerns to find ears and eyes to be considered.

Relying on self-reports to judge whether bureaucrats take citizen concerns into account is problematic. The divergence between studies that rely on self-report measures and those that rely on observational data suggest that when bureaucrats can self-report they might overstate their engagement with input from citizens. Meanwhile observational studies likely under-report on the influence of citizen engagement on policy as they cannot capture individual responses to citizen input. Further, citizen input might find its way into policy design and implementation only gradually. Existing studies have struggled to account for longer-term and slow acting change.

This study addresses some of these methodological shortcomings by using a behavioural measure and an experimental design in which how citizen input is presented to bureaucrats is randomised (see also Table 1 for an overview). The study pre-registered two definitions of the outcome. A scale variable, which is zero when the person did not open the email, one when they opened it, two when they also clicked on the link and three when they spent more than 1 minute exploring the information on the web-pages. Browsing time was recorded as active browsing for ≥ 1 min of cursor activity, scrolling or clicking. This time frame was chosen as it provides for a time window that is (i) long enough to distinguish it from simply opening a tab and navigating away from it and (ii) long enough to navigate to relevant areas and process keywords relevant to one’s area of work. Whilst more granular measures of browsing time might have added to the analysis, this was not technically possible when the trial took place.

The scaling was chosen to reflect the assumption that the more steps a bureaucrat completes, the more access responsiveness is given. Since every step requires additional effort, it is also implied that the more steps a bureaucrat completes, the more motivated they are. Using a scale meanst that distances on the scale are chosen somewhat arbitrary. Therefore, I also converted the scale to Z-score as a robustness check and for improved interpretability of the analysis. In addition, I also ran an ordered logit on the scale variable. I found no difference in results (see online appendix).

In addition, a binary outcome was registered to capture less granular measures of engagement, in the case that the browsing tracking was not deployed correctly. It is set to one when a person clicked on the link contained within the email and to zero when they do not (and thereby cannot get on the webpage containing the citizen input) (Table 2).

OPEN IN VIEWER

Table 2.

Overview of outcome variables.

Outcome	Variable definition
Engagement (scale)	Yi in [0,1,2,3], where
	Yi = 0 when email not opened,
	Yi = 0 when email not opened,
	Yi = 2 when opened link clicked
	Yi = 3 when opened and
	clicked report read
Engagement (binary)	Yi in [0,1], where
	Yi = 1 when link clicked,Yi = 0

The outcomes observed immediately after treatment administration regard access responsiveness: whether bureaucrats actually look at and read the information provided by citizens. Policy design and implementation decisions however take place over several days and weeks. Literature on bureaucratic agencies typically looks at budget allocation as a measure of discretion and independence from political mandate (Ting, 2001). Those that investigate the influence of citizens on budgets only find very limited effects (Bould- ing and Wampler, 2010; Guo and Neshkova, 2012; Cohen et al., 2016). The online appendix includes an analysis of budget changes that the LA under study conducted following the citizen consultation. Congruent with expectations, changes are almost entirely explained by the availability of central government grants. The results are thus not discussed further in the main part of the paper.

Participants and Randomisation Strategy

Participants of the field experiment were employees of the local authority. A total of 16,013 individuals are registered in the local authority’s HR database. The study aimed to target bureaucrats who were – albeit to varying degrees – involved in the design and planning of local policies. This meant that 3,498 staff were removed from lists as they worked in unrelated functions – for example drivers, cleaning and catering services – and did not use email on a regular basis for their work. Another 4,988 were on short-term, probationary or casual contracts, meaning that they were less likely to be involved in planning activities: either because they had joined only recently (probationary contracts) or were hired to complete already pre-specified tasks (short-term and casual contracts). Such contract-types also had a higher likelihood of having out-of-date email addresses. Therefore, these contacts were removed from the list before randomisation. Four contacts of senior executives were added as they had not been included in the original email list obtained from HR. The final list of employees eligible for participation in the trial thus consisted of 7,532 employees.

The study used blocked (stratified) randomised assignment. The list of 7,532 eligible bureaucrats was randomised into four arms with three stratification variables (blocks) of two levels each: (i) managerial level (manager, no manager), (ii) working in central planning versus service functions – Lipsky (1983)’s street-level bureaucrats and (iii) previous exposure to the citizen engagement effort (previously exposed, not exposed). The stratification variables were chosen based on the expectation that they would influence treatment effects. Note that no other covariates were used.

It is possible that managers react differently to the intervention compared to people with non-managerial roles. Managers are tasked with designing team strategies to achieve organisational goals; they might therefore face greater pressures to incorporate citizen input than persons who are less involved in planning activities. However, pay grades did not necessarily respond to the degree of responsibility awarded to a person. Following discussions with the authority’s HR team, I determined that role descriptors provided a better proxy for managerial responsibilities.

People were classified as having a managerial role if their role descriptors included any of the following words or abbreviations: ‘manager’, ‘senior’, ‘snr’ and – unless it was paired with ‘assistant’ – the words ‘executive’ and ‘exec’. The data set was manually checked for other abbreviations, in order to assure that these captured all markers of managerial level and seniority. A total of 2,194 persons were of managerial rank (29% of the eligible contacts). Whilst gender was not recorded in staff data, titles (‘Mr’, ‘Mrs’, ‘Ms’ and ‘Miss’) were used to infer gender. Only for four cases the title – ‘Dr’ – had no clear gender. For these four cases, trial partner staff identified whether the first name was male or female. Based on their first names, all four cases were classified as female.

People employed in more local and frontline services (street-level bureaucrats) might react differently to input from citizens compared to those working in more centralised design and planning functions. They might be more inclined to engage because they are more exposed to citizens in their everyday work and therefore perceive citizen opinion to be more relevant to their work. On the other hand, they might be less inclined to explore such information because they already feel they are informed enough through face-to-face interactions.

Based on their registered location of work, I created a binary indicator for working for central functions or more peripheral and front-line functions. The local authority’s strategy, budgeting, planning and other central functions were all located in the county hall at the time of the trial. A total of 1,491 people (20% of the sample) worked in this central location whilst 6,073 worked in other functions across the local government area.

Most local governments manage citizen engagement exercises internally. This was also the case with the local authority taking part in the experiment.

It is therefore expected that several people will have preliminary information about the citizen engagement exercise. Since they were involved early on, they might be more invested in the cause and thus more likely to respond to the treatment.

To the contrary, they might be less likely to engage with further input, because they feel already informed enough. A total of 54 persons had been exposed to information pertaining to the citizen engagement exercise before the trial (0.71% of the sample).

Table 1 shows the number of observations allocated to each condition by block. I use the R-package randomizer to implement this block randomisation. For the full randomisation code, please refer to the supplementary online materials.

All stratification variables used in this trial were categorical. Chi-squared tests showed that – as expected – their distribution did not differ significantly between experimental conditions (Table 3). Randomisation was blocked by managerial level, central versus peripheral work location and previous exposure. Gender was not included in the blocked randomisation but as chi-squared tests illustrate, the difference between the proportion of women and men allocated to each of the treatment arms did not differ from chance (see Table 3).

OPEN IN VIEWER

Table 3.

Balance checks – main treatments.

	Control, %	Treatment 1 (Shortened), %	Treatment 2 (Symbolic), %	Treatment 3 (Value), %	P-value
Blocking variables
Manager	29.03	28.96	28.96	28.96	1.000
Central	19.67	19.75	19.74	19.73	0.999
Prev. exposed	0.74	0.69	0.74	0.69	0.995
Gender
Male	30.41	29.01	30.37	30.41	0.731
Female	69.59	70.99	69.63	69.59
Directorates
Child
Econo	29.03	31.02	30.65	28.93	0.617
Finan	14.38	12.81	14.82	14.44
Health	5.45	5.66	5.35	6.24
Highw	17.77	17.47	17.68	16.87
Trans	5.82	4.71	4.76	4.65
Priority Areas
No priority	71.02	68.98	69.24	71.07	0.33
Priority	28.98	31.02	30.76	28.93

The type of directorate and whether citizens identified the directorate as a priority area could affect the responsiveness of bureaucrats. As additional balance checks show, randomisation allocated bureaucrats of different directorates, with different citizen priorities attached to them, with equal probabilities into different experimental conditions. Observed responses are thus statistically independent of an affiliation to a directorate and area of priority (see Table 3).

Using MailChimp’s ⁶ recording system for ‘hard bounces’ – failed attempts of delivery after seven trials – we can assess whether emails were successfully delivered at the same rate across experimental groups. Emails can bounce due to out-of-date email addresses or technical issues. The email client used for this study captures such bounces. One percent of participants were never-takers because the emails bounced (see Table 4). Because of the low non-compliance rate, I will not estimate the complier average causal effect (CACE) (Figure 1).

OPEN IN VIEWER

Table 4.

Compliance checks.

Compliance		Treatment		Sig.
	Control	Shortened Symbolic	Value	p-value
delivered (no)	99.95%	98.09% 98.99%	97.62%	0.001
bounced (yes)	0.05%	1.91% 1.01%	2.38%	0.001

OPEN IN VIEWER

Figure 1.

Engagement, defined binary: no engagement versus meaningful engagement; by experimental condition. Note: Grey dots indicated individual treatment observations. The darker the area, the more dots are overlaid, meaning that darker areas are indicative of more observations whilst lighter areas indicate fewer observations. The estimated treatment effects and their confidence intervals are overlaid on the raw data points.

Percentages for recorded actions by treatment are summarised below in Table 5 and Figure 2. It is immediately evident that more bureaucrats in the two behavioural treatment conditions engaged with citizen input than in the control and the reduced information treatment. Whilst more people who received the shortened email text clicked on the link than in the control, almost none of them spent time on the website. This suggests that low levels of engagement are unlikely solely due to information overload. Inferential analyses in the next section unpack this further.

OPEN IN VIEWER

Table 5.

Outcomes by experimental group.

	Control	Treatment 1 (shortened)	Treatment 2 (symbolic)	Treatment 3 (value)
Not opened	68.54%	84.75%	58.52%	83.71%
Opened only	30.20%	0.05%	24.50%	0.00%
Opened + clicked	1.27%	12.23%	2.80%	1.53%
Opened + clicked + browsed info on web-page	0.00%	2.96%	14.18%	14.75%
N	1,891	1,889	1,890	1,891

OPEN IN VIEWER

Figure 2.

Estimated treatment effect of each experimental condition on engagement. Note: Grey dots indicated individual treatment observations. The darker the area, the more dots are overlaid, meaning that darker areas are indicative of more observations whilst lighter areas indicate fewer observations. The estimated treatment effects and their confidence intervals are overlaid on the raw data points. Compared to the control, both the value and the symbolic treatment significantly increase engagement compared to the shortened version of the email. The symbolic treatment leads to greater rares of engagement than the value treatment. The effect of the shortened message is statistically indistinguishable from that of the control.

Estimation Strategy: Treatment Effects on Engagement

As pre-registered, I use an OLS regression with fixed effects for blocks and heteroscedasticity-consistent (HC2) standard errors to estimate the average treatment effects and repeat the same analysis using a logit model.

Treatment Effects on Engagement

In line with the pre-registered analysis plan, I present estimates of treatment effects and significance levels using ordinary least squares (OLS) regressions with fixed effects for blocks. ⁷

Looking at engagement as a binary variable – whether bureaucrats clicked on the link to learn more or not – the shortened, value and symbolic treatments all increased engagement compared to the control. Results show that 15% of people in the shortened treatment reacted whilst 16% did in the value treatment. Seventeen percent reacted to the symbolic treatment. These differences are significant at p<0.001. Differences between treatments are however not significant.

However, if we use the scale measure that accounts for the different degrees of effort expanded – from opening, clicking, to actually spending time browsing the information – we see that the value treatment performed significantly better than the shortened message. Using this second pre-registered outcome – a more granular measure – differences become clearer (see Table). Fifteen percent of people in the value treatment read the information whilst only 3% of those who received the shortened message did.

The symbolic and value treatments cause higher levels of engagement (Table). Compared to the control, both the value and the symbolic treatment significantly increase engagement compared to the shortened version of the email (see Figure 2). The symbolic treatment leads to 5% more bureaucrats reading the additional information compared to the value treatment (p<0.001). The effect of the shortened message is statistically indistinguishable from that of the control. ⁸

Notably, there are considerable differences between opening rates between the symbolic treatment condition and the remaining experimental conditions. I thus also calculate the effect of each treatment condition compared to the control, conditional on opening using the Lee bounds estimation procedure (Lee, 2009). Lee bounds take into account attrition at the preceding stage and trim bounds on treatment effects accordingly. As Table 6 summarises, the effect of the value treatment, conditional on opening, amounts to a 91% increase in reading rates. The effect of the symbolic treatment ranges between a 13–45% increase and is thus not distinguishable from the effect of the shortened treatment.

OPEN IN VIEWER

Table 6.

Effects on reading, conditional on opening: Lee bounds for effect sizes.

	Upper/Lower	Coef	CI Low	CI High
C vs Short	Lower	0.19	0.15	0.24
	Upper	0.19	0.15	0.24
C vs Symbolic	Lower	0.13	0.05	0.22
	Upper	0.45	0.39	0.51
C vs Value	Lower	0.91	0.87	0.94
	Upper	0.91	0.87	0.94

In terms of participant characteristics, working in the central planning unit of the local authority is associated with an increased likelihood that a bureaucrat engages with the input (see Figure 3A). Those who were involved in the data collection process also engage more. However, the proportion of participants who were exposed to these processes prior to the trial is very small, meaning that estimates are likely to be imprecise and should therefore be interpreted with caution. What areas bureaucrats work in was also related to their likelihood to engage with input. People working in child and school services were the least likely to engage and people working on strategic functions – service planning, digital government and the transformation programme (corporate restructuring, strategy and budget cuts) were the most likely to engage (see Figure 3D). People working in finance and other planning-heavy departments such as highways and environment were also more likely to engage with input than people working in departments related to schools and child care.

OPEN IN VIEWER

Figure 3.

Estimated treatment effect of each experimental condition on engagement by type. (A) Area of work; (B) Managerial level; (C) Gender; (D) Directorate. Note: Figures A–D show average treatment effects and confidence intervals (dots and whiskers) superimposed on scattered raw data points, broken down by demographics. Heterogeneous treatment effects can be observed for the symbolic reward and the value conditions for bureaucrats working in different areas (Figure 3A) and directorates (Figure 3D). Those working in central, policy-planning as opposed to street-level units react more strongly to treatment interventions. The difference is particularly pronounced for the value-based treatment message, albeit the effects are positive and significant for both types of bureaucrats. Those working in financial planning (Figure 3D) and the transformation department – a department specifically tasked with driving changes related to the consultation underlying this field experiment – responded more strongly to the symbolic and value treatment interventions. Bureaucrats working in the transformation department also responded more strongly to the cognitive-load reducing (shortened) treatment.

To gather additional information on information use, I ran a follow-up survey two months after treatment administration. Turnout was low (5%, N = 339) and results therefore need to be interpreted with caution. Self-reported use of information broadly reflected the ordering of treatment effects: Among respondents who had received the control email and answered the survey, 20% reported that they had used the information to inform their work. This compares to 31% (symbolic) and 38% (value) for the behavioural treatment groups. More information on the survey and its results is available in the online appendices.

Limitations and Areas for Further Investigation

The citizen consultation underlying the study was very high profile: advertisements were ubiquitous and visible for several months across the local authority. The consultation included focus groups, surveys, debates and meetings with different stakeholder groups. On the one hand, this could have led bureaucrats to take the initiative more seriously. On the other hand, if bureaucrats might have taken issue with the low level of deliberation provided by these activities [see Fung (2006)’s hierarchy of participatory initiatives]. It is plausible that this subset of bureaucrats who did not see the consultation living up to standards of participatory governance could not be motivated to engage with its results as they regarded it as illegitimate. It would therefore be instructive to test similar interventions in a context where information is based on more intensive forms of participatory governance initiatives.

Leadership and organisational culture might also influence to what extent the same patterns could be observed in other studies. In this study, bureaucrats were markedly resistant to change. Meanwhile, the senior local authority executive who signed off emails to bureaucrats was well-regarded, which might have alleviated some of the negative effects of low staff morale. The observed baseline and rates of post-treatment engagement might differ in other contexts. It is possible that if stronger treatments would have been used, larger effects would have been observed. A simple ‘thank you’ is perhaps the smallest symbolic reward one could award. If larger rewards such as employee recognition or award ceremonies were used, the effect might have been larger. Another important avenue for future investigation are long-term effects. If value-based communication and small, verbal rewards are not backed up by recognising and rewarding the work of bureaucrats who display desired values and exert additional effort, it is likely that such interventions will soon lose their effect; they might even decrease engagement as lack of follow-through encourages cynicism.