Comparison of COVID-19 Behaviour Change Campaign Ads in Tanzania: Results From a Randomised Controlled Survey-Based Experiment

Abstract

Background

In circumstances where vaccine hesitancy is high, like in Tanzania in 2021, measures to control the spread of COVID-19 infection through non-pharmaceutical interventions, specifically mask-wearing, hand-washing and physical distancing, become crucial. This study was undertaken to inform the development of a context-adapted communication campaign to control COVID-19 in Tanzania.

Focus of the Article

The study examines the effectiveness of three different behaviour change campaign ads created as part of the same control effort, and two of them by the same creative process, with the same creative team, at roughly the same time, in altering behavioural propensities for relevant target behaviours, including the washing of hands, the wearing of masks and social distancing, and key participant attitudes, such as the sense of obligation to others.

Research Question

The main research question was: what are the effects of behaviour change campaign ads on propensities for engaging in COVID-19 target behaviours? We also examined behavioural attitudes and consumer appreciation of the ads (believability, likeability, relevance and surprise). The purpose of the study was to provide empirical evidence on the effectiveness of different messaging logics, in order to help improve future campaigns.

Methods

We designed a controlled survey experiment where participants were randomly assigned to receive one of the three campaign ads. These were called: the “Setting” campaign, which was designed using a theory-based Behaviour Centred Design (BCD) approach and produced with local partners; the “Password” campaign, which was the result of a substantial investment by a major international consortium; and the “Balance” campaign, a local adaption of “Password”; as well as an educational public service announcement, which served as an active control. An SMS survey was administered online to 2080 participants in Tanzania in 2022 to evaluate the potential effectiveness and consumer appreciation of the campaigns.

Results

The results showed that the most costly campaign, “Password”, did not perform better than the other campaigns on any test. In particular, “Password” was not more effective than the theory-based “Setting” campaign on any single behavioural indicator or exposure variable. The “Setting” campaign was more effective than other ads on the expected qualities of appreciation (surprise, believability). However, the educational announcement achieved higher average response levels than all of the narrative-based treatments on measures of consumer appreciation, such as likeability, believability and relevance.

Recommendations for Practice

This study supported the proposition that a campaign specifically designed to elicit particular kinds of psychological responses could do so. It showed that a theory-based campaign, produced locally and at low cost, can compete on standard marketing values with high-quality creative processes and production values.

Keywords

behaviour centred design COVID-19 social marketing communication campaign randomised controlled trial

Introduction

The severe acute respiratory syndrome coronavirus 2 pandemic (hereafter called COVID-19) reached Tanzania in March 2020. Since then, the Tanzanian Government has taken actions to help stop the spread of the virus, including school closures, travel restrictions from affected countries and quarantine for incoming travellers. Vaccine hesitancy and resistance has been a significant problem in Tanzania. President Magafuli, who was a noted opponent to the use of COVID-19 vaccines, died in March 2021, and was succeeded by Samia Suluhu, who was more accepting of vaccines. Results from surveys suggested that around half of the adult population held beliefs and attitudes that ran counter to uptake of the vaccine, and nearly half said they would not take it, due to concerns over safety, side effects, and feeling they did not need it.¹

There are different approaches to developing social marketing campaigns. A recent review indicated the importance of incorporating behavioural theory and using local knowledge in the design of campaign materials (Green et al., 2019). This paper aims to help fill knowledge gaps around strategies for effective infection prevention in disease outbreak settings. The overarching goal of the study was to inform decisionmakers about which components of messaging campaigns were likely to be effective for improving safety and hygiene practices to prevent severe acute respiratory infections, and thus help others to design more effective communications in future. The paper therefore provides a case study of alternative campaign messaging options, rather than a test of a specific social marketing theory. We incorporated different theoretical strategies in the design of the test materials and aim to explain our findings via standard social marketing variables.

We compare messages generated using different creative processes but with the common objective to prevent COVID-19 transmission, via a survey which was delivered to participants using a randomised controlled trial (RCT) design. The outcomes of interest were ratings on a variety of behaviour change indicators and consumer appreciation qualities. The specific evaluation questions were:

• What are the effects of campaign ads produced using different creative approaches on indicators of behaviour change potential?

• What are the effects of the ads on attitudes (behavioural intention) and standard consumer appreciation measures including believability, likeability, relevance and surprise?

The results of this evaluation were used to inform the development of context-adapted activities and materials for the Tanzanian government’s COVID-19 behaviour change communication campaign. Subsequent sections of this paper present a description of the various creative processes and resulting ads, study objectives, the methods and results. The final sections discuss the findings and conclude.

Background and Description of the Ad Campaigns

Social marketing tends to be profligate in its use of theory, there being a lack of consensus about which is most appropriate in a given situation (Lefebvre, 2000). Many of the theories used are adopted from public health, such as the health belief model (Strecher & Rosenstock, 1997), social cognitive theory (Bandura, 1986), and theory of planned behaviour (Ajzen, 1991). In all of these theories, human intentions to act arise from perceptions about the action, perceived norms and self-efficacy. Therefore, the likelihood of performing a behaviour is the result of a complex calculation involving attitudes about the action itself (for example, whether it is good or bad), the perceived ability of the individual to undertake the action successfully, and expectations about the attitudes and behaviours of others (Wolff, 2021). For example, a study to understand whether the theory of planned behaviour could explain variation in COVID-19 protective behaviours (hand-washing, mask wearing and social distancing) showed several of its variables were significantly correleated with reported behaviour across two European countries (Wollast et al., 2021).

Rarely is it the case, however, that different intervention strategies are compared for their ability to elicit behaviour change. The present study aims to do just this; it compares three different potential campaign strategies, developed using different creative processes by different teams, but roughly at the same time, to address the COVID-19 pandemic in Tanzania. Each strategy is represented by a one-minute broadcast-quality television advertisement that encompassed the central insight from three different approaches to developing the campaigns. All campaigns reflected circumstances on the ground, based on field research, and were motivational, as a consequence of using design processes and a number of in-country and international creative partners and media production teams. We describe the logic of each campaign in turn in this section. Screenshots of the campaign ads are provided in Appendix 1.

Intervention 1: The “Setting” Campaign

A behavioural science-based campaign for COVID-19 control was developed by a team of creative agencies from India and academics from UK universities. This was based on the belief that a theory-based campaign would be more effective, even against a global pandemic, when most existing messaging was simply fear-based. A human and behaviour-centred design process was used to generate ideas through a succession of workshops, which were conducted on-line. The process followed the program development steps of Behaviour Centred Design (Aunger, 2020; Aunger & Curtis, 2016). This approach combines behavioural science and design thinking processes to achieve behaviour change. Key tenets of the approach are to design an intervention which prompts audiences to re-evaluate their behaviour and change performance in specific settings. The drivers of human behaviour are presumed to be evolved emotions, drives and interests that arise in different parts of the brain.

A concept central to BCD is that of a “behaviour setting” (Barker, 1968). A behaviour setting is a coordinated set of factors that circumscribe performance of everyday actions to achieve a common objective. Examples include going to a restaurant (to eat), going to math class (to learn math), shopping in a store (to purchase desired objects), attending a religious service (to satisfy spiritual needs), or standing at a bus stop (waiting to catch a bus). As with all behaviour settings, these require people to play various roles (e.g., waiter, teacher, driver, customer) supported by infrastructure (such as buildings or buses) and facilitated by objects, called “props”, like plates, glasses, tickets, money, books, etc. A behaviour setting can only be executed successfully to achieve the desired outcome in a particular context when all of its constituent parts fulfil their roles. This concept was used throughout ad development as the central theoretical insight.

The contrivance underlying the ad is that the infrastructure and props stop working automatically when any actor does not play their role properly. Mechanical processes enforce the setting norms in people: the setting goes “on strike” until people conform and any miscreants are corrected. In this case, that means getting people to conform with the norms around the target behaviours. So, when a person enters a public setting without wearing a mask, or washing hands, everything stops working until the miscreants remedy what they have been remiss about. All necessary conditions have to be met for the setting to resume – in other words, everyone should be wearing a mask, practising handwashing, and social distancing. Thus, the setting breaks down for every person in the setting, not just one person – which demonstrates that the setting is a unified whole, that there are mutual dependencies between roles – and increases the social pressure on the human participants to find a solution.² The human motives this approach attempted to activate were ‘affiliate’ and ‘justice’.

Intervention 2: Global “Password” Campaign

The United Kingdon’s Foreign, Commonwealth and Development Office (FCDO) and Unilever formed a coalition, the Health Behaviour Change Coalition (HBCC), to manage their funding efforts in response to the COVID-19 pandemic. The HBCC management committee, having funded 23 different organisations to produce campaigns in 37 countries, recognized that not all of these organisations were expert at behaviour change communications. Since media were needed in many cases (given restrictions on interpersonal contact due to the pandemic), it was decided to put part of Unilever’s funding contribution and long-term associations with well-known, international advertising agencies to good use, and to go through a standard marketing process to choose the agency with the best creative idea. The winning agency was subsequently funded to produce a COVID-19 control campaign that could be used by all coalition partners.

Considerable effort went into the development of the creative brief, which reflected concerns about its global applicability and timeliness, given the phase of the pandemic at that time. The brief was based on an insight from research done by a Kenyan member of the coalition management: “I’d rather die of COVID than of starvation (from not being able to go to work)”. That is, people were more afraid of the economic hardship derived from the political response (i.e., lockdown) than the health risks coming from the pandemic itself. This was at the time when the concept of the “new normal” and “build back better” were gaining world-wide currency. The insight in the brief was that “consistently practicing a few simple behaviours gives me the peace of mind to go forward with my life (earning a livelihood while protecting the health of my family and friends) in this new COVID reality.” The resulting campaign was built around the idea that protecting oneself from COVID by practicing the well-known behaviours would enable people to get back to work, epitomized by the ‘sounds of life’, such as the sound of a motor for a taxi-driver, or clanging metal for a factory worker.

Due to pressure from the HBCC, however, greater emphasis on the target behaviours was felt to be required and the agency was asked to produce a new concept. The idea that came forward was that people could demand of each other that they practise the target behaviours through a reminder at particular junctures, asking the question “What’s the password?” – meaning, “What do you need to do now for all of us to stay safe?” One of the insights from pre-testing was that people liked seeing many others having to practise the behaviours, so the range of situations depicted included entering a place of business, coming home from shopping, and leaving the home. A number of versions of the final execution were produced for different continental groups, given the global coverage of the partners.³

Intervention 3: “Balance” Campaign

By 2020, many people in Tanzania believed that COVID-19 was at low prevalence, suggesting that Tanzanians had won against COVID-19 by adhering to safety measures while continuing business as usual (since there was no lockdown). This situation led to the insight to use the sense of victory in inspiring people to continue practicing the behaviours to sustain that victory. The objective was for the target audience to feel inspired to keep practicing the three hygiene behaviours as the secret of their victory against COVID-19. The slogan developed was “Siri ya ushindi” (“the secret of our success”), a hint that keeping doing the right things was key to maintaining the desired behaviour during the pandemic.

Just as this campaign was about to be rolled out, however, the Project CLEAR team became part of a successful bid for HBCC funding, and was charged with also developing a campaign consistent with the HBCC Global COVID-19 password campaign described above. Since the political circumstances in Tanzania were such that COVID-19 could not be mentioned explicitly, and the idea of “success” was already present, the task became one of fitting the “Password” campaign to these local circumstances, and give it greater motivational power. A creative brainstorm suggested the idea that the secret itself was a “balance” between work and safety (consistent with the original insight of the Password campaign). Balance is the way in which you can keep safe while still working, avoiding the dangers of illness and death from COVID-19 on one hand, and hunger and poverty on the other. Tanzania’s “secret” is that Tanzanians have maintained the proper balance in life. The slogan thus became that “Siri ya kushindi ni ku balance” (“the secret to success is to keep the balance”). To keep winning, you have to keep practicing the prescribed behaviours, thus creating balance. In this interpretation, the key element is the idea that doing some simple things is the secret key, the real enabler of COVID-19 control. The concept says that it is easy to stay safe by just doing these behaviours, which allows you to get back to normal life. However, the job is not finished – the virus can always come back – so everyone must keep doing the right things. As fatigue set in with respect to doing the target behaviours, the campaign needed to reinvigorate the population. Those who continue to perform these behaviours are morally good (protecting others) and should be granted higher social status.⁴

Control Campaign ad

A fourth ad about COVID-19 was used as an active control.⁵ This was an educational advertisement produced as a public service in Swahili.

Approach

Study Design

The study is an individual-level randomised controlled trial (RCT) to estimate the effect of three different behaviour change campaigns on reported behavioural intentions. The study comprised four arms: three arms which included one of the COVID-19 campaign advertisements, and an active control arm that included a knowledge-based COVID-19 elimination campaign advertisement.

Individuals were self-selected from an existing list of mobile phone numbers used for marketing research purposes by the data collection partner, GeoPoll Ltd., an international consumer research company. Participants eligible to participate in the study were those over 18 years of age, living in the study country and registered in GeoPoll’s user consumer database in Tanzania. Participants needed to use a smartphone to access the intervention ad. Compliant individuals were subsequently randomised to view one of the four ad videos. They were sent a short messaging service (SMS) message containing a link to a website which contained the intervention material, to which they were exposed, followed by a suite of questions concerning that intervention.

The survey instrument was administered by GeoPoll Ltd. in March, 2022. The primary study outcomes were responses to hypothetical scenarios relevant for COVID-19 control in which participants could choose their preferred actions. The behavioural targets of these campaigns were: physical distancing in public places; handwashing with soap at relevant occasions; and wearing a mask when in proximity to closed-in groups.

A total of 2080 participants were recruited, 520 in each arm (3 treatments and 1 control). We assumed that each intervention was associated with at least a 10 percent difference in the primary outcomes compared to the control arm, and that the target behavioural science campaign was associated with a 10 percent difference compared to the other interventions. Assuming, conservatively, that the primary outcome had a prevalence of 50 percent, then identifying a 10 percent difference to any of the interventions with 90 percent statistical power would require a minimum sample size of 400 people per arm. We increased the sample size to 520 per arm to allow for incomplete data (20% losses to follow-up). Table 1 presents the questions and possible responses pertaining to the primary and secondary outcomes in the order in which they appeared in the survey (the full survey in English and Swahili is in Appendix 2).

Table 1.

Outcome Questions and Responses.

Outcome	Survey questions	Responses
Consumer appreciation (secondary research question)	To what extent is the video surprising to you?	Very much
	To what extent is the video believable?	A little
	To what extent does the video have meaning for you (relevance), in your life?	Not at all
	To what extent is the video likeable?
Attitudes (secondary research question)	Do you believe seeing the video would encourage people to increase handwashing, keeping distance, and wearing a mask in public?	Very much
	I feel it is my personal duty to protect others from possible infection, even at a cost to myself.	A little
	Seeing this video makes me want to avoid going to large indoor gatherings like funerals, church services and cinemas.	Not at all
	Seeing this video makes me want to wear a mask whenever I go out in public.	Strongly agree
	Seeing this video makes me want to wash my hands with soap after I go out in public.	Agree somewhat
		Do not agree
Behavioural propensity (primary research question)	Imagine you are in a queue to pay in a mall shop, the person behind is not wearing a mask. Would you to tell them to put on a mask?	Very likely
	Imagine you are in a queue to pay in a mall shop, the person behind is close by. Would you to tell them to move away?	Somewhat likely
	Imagine you have been trying on clothes in a mall shop. How likely are you to use the hand sanitizer by the door as you leave?	Not at all likely

A small, single-purpose study of responses to the slogans associated with each campaign was also undertaken by GeoPoll Ltd. in June 2021, as part of a regular data collection plan that was part of the government’s monitoring of COVID-19-related beliefs and practices. Data were collected by a mobile online survey conducted by the Kenyan office of GeoPoll Ltd., from their database of potential Tanzanian respondents. This slogan study was a regular panel survey with a sample size of 1122.

Data Analysis

Statistical comparisons were made between Likert-scaled responses to questions concerning surprise, likeability, relevance and intention to engage in the target behaviour. Primary study outcomes were:

• Scenario-based behavioural propensity indicators for the relevant target behaviour (mask wearing, maintaining distance, hand-washing with soap).

• Whether the participant agreed to receive further information about COVID-19 vaccine availability.

Secondary outcomes were:

• Consumer measures of appreciation (believability, likeability, relevance to consumer’s life, surprise).

• Measures of attitudes including the intention to perform the relevant target behaviour in a particular behavioural scenario (social distancing, face-mask, handwashing), the likelihood that others will perform target behaviour (encouragement to wash hands) due to campaign exposure, and the sense of obligation to others to protect public health.

Descriptive statistics were prepared on participant socio-demographics, secondary outcomes and understanding of intervention and behaviour change indicators. For continuous variables these included the mean, standard deviation, median, interquartile range and the number of observations. Categorical variables were summarised as counts and proportions.

Analysis aimed to determine whether campaigns differed significantly in their ability to produce positive responses to behavioural propensity indicators. It also tested for differences in the means of the consumer appreciation variables between intervention campaigns and with the control group, using the same approach. The primary and secondary outcomes were compared across arms using prevalence differences with 95 percent confidence intervals.

Bivariate analyses using chi-square tests were used to explore the relationships between demographic characteristics, programme exposure, and primary outcomes. Multivariate analysis used linear probability models. Individual outcome variables were measured across three ordinal categories (e.g., for behaviour scenarios these were: 0 = “not at all”, 1 = “a little”, 2 = “very much”; responses to attitudes and consumer appreciation questions were also coded similarly from zero to 2). Indexes for primary and secondary outcomes were calculated as the sum of the individual outcomes under each category. For example, for the primary outcome, the index summed the ordinal responses to questions about mask-wearing, maintaining distance and hand-washing, with possible values ranging between zero and 6. The three regression models were:

{B E H A V I O U R}_{i} = f ({T R E A T M E N T}_{i j}, {C O V A R I A T E S}_{i})

(1)

{A T T I T U D E}_{i} = f ({T R E A T M E N T}_{i j}, {C O V A R I A T E S}_{i})

(2)

{A P P R E C I A T I O N}_{i} = f ({T R E A T M E N T}_{i j}, {C O V A R I A T E S}_{i})

(3)

where

{B E H A V I O U R}_{i}

(value range 0–6) is the scenario-based behavioural propensity index (primary outcome) for each participant i,

{A T T I T U D E}_{i}

is the behavioural intention index (value range 0–10),

{A P P R E C I A T I O N}_{i}

is the consumer appreciation index (value range 0–8),

{T R E A T M E N T}_{i j}

represents the campaign exposure variable j randomly assigned to individual i, and

{C O V A R I A T E S}_{i}

are socio-demographic variables (sex, age, highest education level, occuption and location). Standard errors were clustered at the administrative level.

We used the standard threshold of p < .05 to determine if the analyses suggested that the results were significantly different from those that would be expected if the null hypotheses that there were no differences between treatment effects for each campaign were correct. No adjustments for multiple testing were done and analyses reported were pre-specified, with the exception of the regression analysis, which was conducted post hoc. Informant responses that included missing data were excluded by Geopoll from their databases. Informants were excluded if their responses showed no variability (e.g., the first option among potential responses was always selected).

Research Ethics

Ethical approval was obtained from the National Institute of Medical Research (NIMR) in Tanzania (Approval No. NIMR/HQ/R.8a/Vol. IX/3863) and the Ethics Review Committee at the London School of Hygiene and Tropical Medicine (LSHTM) (Approval No. 26464) prior to commencement of the study. The study design and data analysis plan were preregistered with the Centre for Open Science.⁶ The data were hosted, updated and stored in a fully secured database at LSHTM. Any transfer of data only occurred through encryption. Access to password-protected, encrypted databases and data storage drives was restricted to the principal investigators and co-investigators listed in this protocol only. Data will be stored in this location for five years from the completion of the project, and then destroyed.

Informant consent used standard GeoPoll procedures, in conformity with international consumer testing protocols and standards, which were registered via response to an SMS message. Participants in the online surveys were given information about the survey when they clicked on the link in the SMS. Participants were given up to 48 hr after they had opened the link to decide whether they would like to participate in the study or not. Once participants consented by agreeing to the statements in the written consent form, the exposure and questioning commenced. All those enrolled were allocated an identification number. These numbers were used to identify all records relating to the individual. Data were stored in an electronic spreadsheet and were identifiable by number only. Only data from completed surveys were retained by GeoPoll.

Those who completed the survey were compensated 0.5 USD for their time to cover opportunity costs and costs of charging their phones. The money was transferred using M-pesa (electronic money transfers) within 2 hr of completing the interview.

Results

Description of Participants and Balance Tests

From the GeoPoll Ltd. marketing database population, 2610 study participants were randomly assigned to one of the treatment conditions, of which 2080 completed the survey and were included in the analysis (Figure 1). Participants were randomly allocated until there were equal numbers of treated units by group. There was less than one percentage point difference in losses to follow-up across groups.

Figure 1.

CONSORT participant flow diagram.

Nearly half (47 percent) of participants who completed the survey were women, two-thirds lived in urban areas, representing all regions of Tanzania, and around a third each were aged 18–24, 25–34 and 35 or more. It being implemented online, over 80 percent of participants had completed secondary education or higher, and two-thirds were self-employed or in salaried positions. A Kruskal-Wallis test suggested that the samples were balanced by age group (Chi-sq = 0.787, p > .853). There was a slight imbalance by sex in the control group, 60 percent of whom were female, otherwise males and females were equally likely to participate in each group. Other socio-demographic characteristics were broadly similar across campaigns and the control group (Table 2).

Table 2.

Socio-Demographic Characteristics by Study Group.

Participant characteristic		Total		“Setting”		“Balance”		“Password”		Control
Participant characteristic		Count	Percent	Count	Percent	Count	Percent	Count	Percent	Count	Percent
Total		2080	100	520	100	520	100	520	100	520	100
Age group	18–24 Years	736	35	170	33	180	35	186	36	200	38
	25–34 Years	745	36	207	40	198	38	175	34	165	32
	35+Years	599	29	143	28	142	27	159	31	155	30
Sex	Male	987	47	259	50	259	50	259	50	210	40
	Female	1090	52	260	50	260	50	260	50	310	60
	Prefer not to say	3	0	1	0	1	0	1	0	0	0
Region of residence	Arusha	206	10	46	9	57	11	54	10	49	9
	Dar-es-salaam	419	20	90	17	114	22	104	20	111	21
	Dodoma	221	11	62	12	53	10	53	10	53	10
	Kilimanjaro	125	6	27	5	30	6	36	7	32	6
	Mbeya	117	6	30	6	29	6	32	6	26	5
	Mwanza	135	6	37	7	33	6	29	6	36	7
	Other	857	41	228	44	204	39	212	41	213	41
Location	Urban	1403	67	342	66	361	69	331	64	369	71
Location	Rural	677	33	178	34	159	31	189	36	151	29
Highest level of schooling completed	None	33	2	8	2	6	1	14	3	5	1
	Primary	285	14	78	15	61	12	83	16	63	12
	Secondary	888	43	206	40	230	44	208	40	244	47
	University	874	42	228	44	223	43	215	41	208	40
Primary source of income	None	165	8	31	6	38	7	37	7	59	11
	Self-employed	949	46	229	44	247	48	249	48	224	43
	Daily wage	232	11	63	12	63	12	48	9	58	11
	Salaried job	452	22	121	23	115	22	106	20	110	21
	Farming	282	14	76	15	57	11	80	15	69	13

Note. categories may not sum to 100% due to rounding errors.

Results for Primary Outcomes

Table 3 reports the results for the primary outcomes: behavioural propensity indicators for the relevant target behaviour and whether the participant agreed to receive further information about COVID-19 vaccine availability. There were a number of significant differences across the campaigns in their ability to get people to report intentions to behave in accord with pandemic control. Most of these differences focused on the “Balance” campaign ad. This ad was significantly less effective than all three comparators on the mask wearing scenario, the other treatment ads on telling someone else to put on a mask, the “Setting” and control ads on use of hand sanitizer upon leaving a shop, and the control ad on handwashing. None of the campaigns differed in their ability to inspire participants to seek additional COVID-19 information from another webpage, our other primary measure of behavioural consequences from exposure.

Table 3.

Significance Tests for Primary Outcomes.

Outcome		Pair-wise tests by study group
Outcome		Balance	Password	Setting	Control
Avoid large indoor gatherings	Balance
	Password	0.416		0.837	1
	Setting	0.543
	Control	0.416		0.837
Wear mask in public	Balance
	Password	0.032**		1	0.930
	Setting	0.032**
	Control	0.039**		0.451
Wash hands with soap	Balance
	Password	0.107		0.671
	Setting	0.234
	Control	0.052	0.742	0.451
Telling others to put mask on	Balance
	Password	0.043**		0.945	0.449
	Setting	0.050**			0.492
	Control	0.204
Telling others to move away	Balance
	Password	0.293		0.620	0.776
	Setting	0.578
	Control	0.442		0.833
Use hand sanitiser as they leave a shop	Balance
	Password	0.189
	Setting	0.039**	0.453
	Control	0.031**	0.397	0.923
Receive more information about COVID-19	Balance
	Password	0.370		0.424	0.482
	Setting	0.923
	Control	0.846		0.922

Notes. p-values are reported for Chi-squared tests in italics if statistically significant at conventional levels (** indicates significance at p < .05); values appear in the cell associated with the row of the campaign that is more effective (e.g., “Setting” is significantly more effective than “Balance” on use of hand sanitiser); sample size for each pair-wise comparison = 1040.

Results for Secondary Outcomes

Results of the analysis for secondary outcomes are reported in Table 4. These include measures of consumer appreciation (believability, likeability, relevance and surprise), the likelihood of persuading others to wash their hands, and the sense of obligation to others to protect public health. “Setting” was perceived to be the most surprising campaign, although not significantly more surprising than “Password” (p > .1). This was as expected, since the “Setting” campaign was purposefully designed to surprise participants. In contrast, the control advertisement was considered to be the most believable (p < .01). The “Balance” campaign advertisement was also seen as more believable than the “Setting” (p < .05).

Table 4.

Significance Tests for Secondary Outcomes.

Outcome		Pair-wise tests by study group
Outcome		Balance	Password	Setting	Control
Surprising	Balance
	Password	0.048**			0.149
	Setting	0.004***	0.369		0.019**
	Control	0.591
Believable	Balance		0.475	0.040**
	Password			0.180
	Setting
	Control	0.001***	<0.001***	<0.001***
Relevance to your life	Balance		1.0
	Password
	Setting	0.415	0.415
	Control	0.002***	0.002***	0.026**
Likeable	Balance		0.198	0.453
	Password
	Setting		0.592
	Control	0.046**	0.001***	0.006***
Encourage others to handwash	Balance			0.426
	Password	0.645		0.209
	Setting
	Control	0.183	0.384	0.034**
Personal duty to protect others from infection	Balance		0.063
	Password	0.636		0.020**	0.170
	Setting
	Control	0.624		0.335

Notes. p-values are reported for Chi-squared tests in italics if statistically significant at conventional levels (***, ** indicate significance at p < .01, p < .05 respectively); values appear in the cell associated with the row of the campaign that is more effective (e.g., “Setting” is perceived to be significantly more surprising than “Balance”); sample size for each pair-wise comparison = 1040.

The control advertisement was seen as having the greatest relevance to the participants’ life (p < .05). The control was also more effective on the likeability criterion against all the narrative-based campaigns (p < .05). The control was perceived to be more powerful in persuading respondents that other people would engage in the protective behaviour of washing their hands after viewing than the “Setting” campaign (p < .05), but not any others.

The “Password” campaign was perceived to inspire a desire to protect others more than the “Setting” campaign (p < .05). The control appeared to be the most effective on believability, likeability and relevance (p < .05). Among the campaigns, the “Setting” appeared to be most effective, but not significantly so, on meaning and intent-to-behave, and second most effective on surprise and likeability. The “Balance” campaign was not rated highest on any measure.

Regression Results

We estimated linear probability regression models to estimate the effects of the three campaigns on primary and secondary outcome indexes, alongside other pre-existing socio-demographic variables. The results (Table 5) indicated that participants tended to have high behavioural, attitudinal and appreciation index values, but confirmed that “Balance” performed significantly more poorly on behavioural scenarios and attitudes measures. In addition, “Password” performed significantly worse than control on appreciation. The analyses also indicated men were significantly more likely to indicate appreciation of the ads, positive attitudes and intended behaviours than women. Unemployed people were significantly less likely to adopt positive attitudes or behaviour. As age increased, participants were significantly less likely to appreciate the ads or change their attitudes.

Table 5.

Regression Estimation of Primary and Secondary Outcomes.

	Behaviour index		Attitude index		Appreciation index
	Coef.	p > t	Coef.	p > t	Coef.	p > t
Balance	−0.131**	0.028	−0.199**	0.046	−0.227	0.206
Password	0.016	0.760	−0.020	0.859	−0.208**	0.040
Setting	−0.012	0.871	−0.095	0.500	−0.179	0.118
Control^
Age	−0.008	0.115	−0.013***	0.005	−0.013***	0.008
Male	0.119***	0.001	0.321***	0.000	0.635***	0.000
Female^
Primary	−0.114	0.667	0.098	0.807	−0.155	0.702
Secondary	−0.026	0.914	0.431	0.231	0.427	0.147
Tertiary	−0.026	0.917	0.466	0.226	0.289	0.302
No education^
Rural	0.111	0.214	0.124	0.281	0.010	0.916
Urban^
Farmer	−0.097	0.231	−0.001	0.995	0.363	0.012
Unemployed	−0.167	0.058	−0.310**	0.014	−0.078	0.564
Employee	−0.022	0.846	0.050	0.717	0.089	0.517
Self-employed	0.063	0.472	0.138	0.221	0.353**	0.010
Wage labourer^
Constant	5.414***	0.000	8.914***	0.000	5.85***	0.000
Number of obs	2080		2080		2080
Mean (SD) of DV	5.21 (1.18)		9.06 (1.67)		6.11 (1.80)
Root MSE	1.182		1.647		1.756

Notes. ^ reference category; ***, ** indicate statistical significance at p < .01, p < .05 respectively; SD standard deviation; DV dependent variable; MSE mean squared error; standard errors clustered at administrative level.

Slogan Results

The smaller national survey conducted in Tanzania around the same time comparing the slogans associated with these ads showed similar patterns of results (Table 6). The Balance slogan (“The key to success is to keep a balance between safety and normal life”) was less surprising than the other two slogans (p < .10). Password (“The password for taking our world back from coronavirus is ‘Hands-Face-Space’”) was more likeable than Setting (p < .10). Setting (“The world runs smoothly when we all do our part”) was seen as more relevant than Password (p < .10). However, absolute differences in all cases were small. As with the ads themselves, the slogans did not differ in their ability to induce a reported likelihood of behaviour change.

Table 6.

Results of Slogan Survey for Secondary Outcomes.

Measure	(1) Setting slogan		(2) Password slogan		(3) Balance slogan		t-test of differences
Measure	N	Mean	N	Mean	N	Mean	(1)–(2)	(1)–(3)	(2)–(3)
Surprise index	377	0.42 (0.02)	377	0.44 (0.02)	366	0.36 (0.02)	−0.02	0.06	0.07**
Likeability index	377	0.80 (0.02)	379	0.84 (0.02)	366	0.80 (0.02)	−0.04	−0.00	0.04
Relevance index	376	0.92 (0.01)	379	0.90 (0.01)	366	0.91 (0.01)	0.03	0.02	−0.01
Behavioural intention	377	0.93 (0.01)	379	0.91 (0.01)	366	0.90 (0.01)	0.02	0.02	0.01

Notes. measurement scales zero (“Not at all”), 0.5 (“A little”) 1 (“Very much”); standard errors in parentheses clustered by region (25 regions); values in t-tests are mean differences; ** indicates significance at p < .05.

Discussion

For the most part, the campaigns did not differ in their abilities to inspire reported intentions to follow COVID-related behaviours, with the important exception of the “Balance” ad. Whether this is a sign of weakness in the campaign concept, the advertisements, or the character of the survey instrument is difficult to ascertain. However, the context of this kind of exposure can be considered relatively weak – a series of text messages and a one-minute advertisement appearing on the screen of a mobile phone in the middle of someone’s daily activities, during a period when people were likely exposed to a wide range of external messaging about the pandemic.

The “Balance” campaign proved to be least effective on a few behavioural indicators and across multiple dimensions of initial response. The campaign was nevertheless based on extensive formative research and the involvement of the same creative and production teams behind the more successful “Setting” campaign. We suggest that the reason it performed less effectively could be due to the circumstances depicted in the campaign advertisement – with business closures and emphasis on physical distancing even at work – which were no longer relevant at the time at which this study was conducted in 2022. This campaign had been designed for the early experience of COVID-19 in Tanzania, in mid-2020. In addition, the “Balance” campaign ad showed a dust mask being worn at a carpenter’s shop, to protect against flying wood shavings, which may have confused viewers. This also seemed to mirror the degree of emphasis within the “Balance” ad about handwashing, which showed only a very brief moment of handwashing. The control was more effective in persuading people that others would engage in handwashing; given this message was exclusively about washing hands, this result is perhaps not surprising. All of the treatment campaign ads showed handwashing, although with sanitizer and not with soap and water.

There were no significant differences across campaigns in the ability to get informants to report an intention to keep their distance. This may have been because no campaign advertisement showed this behaviour explicitly, or referred to it. The “Setting” campaign advertisement showed people distancing themselves in a queue, but attention was not drawn to this. Thus, there seems to be a literal-mindedness in the pattern of responses to the campaigns: only what was emphasized visually was picked up in the questionnaire responses.

The results with respect to exposure-related variables were quite satisfying. In particular, some evidence supported the proposition that a campaign specifically designed to elicit particular kinds of psychological responses could do so, as the effectiveness of the “Setting” campaign exceeded that of other treatments on the expected qualities (surprise, believability). The “Setting” campaign was designed to be unbelievable—physically impossible things happen in the video—hence the findings were anticipated. An important implication of these results – that the “Setting” campaign is both surprising and unbelievable – is that, to some extent, it runs against marketing standards, which tend to want high believability. In this case, the events depicted in the “Setting” are meant to be shocking in the sense that they go against common physical or technological principles—for example, that machines do not know when you have conformed to some social norm—but to make the point that the (social) world works better when everyone is helping control the pandemic.

An important source of bias in trials using reported outcomes can occur when participants are not masked to intervention status (Savović et al., 2012). The use of an active control rather than a passive one helped ensure that participants were masked as to whether or not they received any intervention. However, the use of an educational campaign as the control may have proven to be an inappropriate choice, as it had qualities the audience preferred to any of the narrative-based treatments in some respects, such as declarative information and directive authority. This was probably because it was the only one which is of an educational nature from an obviously authoritative source. Research in Ghana suggested that social pressure from opinion leaders had a significant effect on COVID-19 related behaviour (Tweneboah-Koduah & Coffie, 2022). People tend to respond to educational compaigns and public service announcements differently to ones that are seen as “branded” (e.g., (Darmawan, 2020)), and therefore self-serving for some private enterprise. Further research is needed to help understand some of the specific mechanisms at play in this study, such as why the public service announcement message was rated as being more likeable than the narrative-based campaigns.

We also note that, while some of the outcomes studied in this paper are similar to those commonly tested in theory-based studies, such as intention in the theory of planned behaviour, our aim was to compare the responses to campaign ads produced using different models of creative development, where the responses were measured using standard marketing variables. Since we were not able to collect data on actual behaviours, we are not able to assess how our results fared in fulfilling the expectations of particular behaviour change theories. However, we can postulate motivating drivers to each of the campaigns, including those not designed using the BCD approach. As noted above, in the language of BCD, the motives thought to underlie the “Setting” campaign were ‘justice’, because of the collective responsibility of all to protect public health, and ‘affiliate’, because of the need to adhere to prescribed behaviours. The main motivating driver for the “Balance” campaign, which appealed to Tanzanian national pride, was ‘status’. In contrast, for the “Password” campaign, which featured people demonstrating they were practising desired behaviours in different situations, the key motivating factor operating appeared to be ‘affiliate’. We posit that, by appealing to a narrower range of behavioural motives, and not adequately incorporating higher order motivating drivers like justice and status, the “Password” campaign was inherently more limited both in its consumer appeal and its ability to alter attitudes and behavioural propensities, all else being equal. Or, to put it another way, we hypothesize that, by appealing to higher order drives, the locally designed “Setting” campaign was able to compete (and out-perform on several metrics) the global advertising campaign. Further empirical research that aims to understand how articulated drives affect behavioural intentions, and subsequently observed behaviour, would provide useful evidence for the design of effective public health campaigns.

Conclusion

This paper sought to take advantage of a rare opportunity to investigate the reasons why some messages ‘work better’ than others at achieving communication objectives. This evaluation reports on a test of the process of designing a behaviour change communication campaign to support national efforts to promote handwashing with soap and other behaviours for the prevention of COVID-19. The existence of several different campaign ads, developed using distinctive creative processes for broadcast in Tanzania by different teams of creative professionals, enabled a randomized comparison of ads with different qualities for their reported effectiveness. A number of statistically significant differences were found in primary behavioural tendency outcomes, although these concentrated on one ad (“Balance”) that was found less effective than the others.

The campaign that was the result of a very large investment by a major consortium, using a multi-stage creative development process led by an internationally renowned creative company, only did better than the other campaigns on one test. In particular, it was not more effective than the theory-based campaign on any behavioural indicator or exposure variable. Further, this latter campaign was significantly more effective at making people feel it was their personal duty to protect others from infection. This provides some evidence that a theory-guided development process, produced locally and at low cost, can compete with high-quality standard advertising processes and production values.

Supplemental Material

Supplemental Material - Comparison of COVID-19 Behaviour Change Campaign ads in Tanzania: Results From a Randomised Controlled Survey-Based Experiment

Supplemental Material for Comparison of COVID-19 Behaviour Change Campaign ads in Tanzania: Results From a Randomised Controlled Survey-Based Experiment by Robert Aunger, Vishna Shah, Hugh Sharma Waddington, and Kaposo Mwambuli in Social Marketing Quarterly

Footnotes

Acknowledgments

Thanks to Joy Masimane from GeoPoll Ltd who provided the participant flow data.

Author Contributions

Robert Aunger: conceptualisation, formal analysis, investigation, methodology, writing - original draft

Vishna Shah: formal analysis, investigation, methodology, writing - original draft

Hugh Sharma Waddington: formal analysis, writing - original draft

Kaposo Mwambuli: investigation, writing - review and editing

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This study was funded by FCDO and Unilever through the COVID-19 Hygiene and Behaviour Change Coalition.

ORCID iD

Hugh Sharma Waddington

Supplemental Material

Supplemental material for this article is available online.

Notes

Author Biographies

Robert Aunger is Associate Professor in Evolutionary Public Health at the London School of Hygiene and Tropical Medicine (LSHTM). He has taught at Northwestern, the University of Chicago and King’s College, Cambridge. He has published books on cultural evolution, the evolution of human behaviour, and ethnographic methods. He is the lead developer of a novel approach to behaviour change called ‘Behaviour Centred Design’. He has helped implement public health projects in water, sanitation, hygiene, nutrition and HIV using this approach in multiple countries on the African and Asian continents. He has also consulted on behaviour change for the World Bank, UNICEF, the Global Alliance for Improved Nutrition, Unilever, Procter and Gamble, Kimberly Clark, and GoJo Industries.

Vishna Shah is a Research Fellow in the Environmental Health Group at LSHTM. In the past five years she has been working on intervention design and delivery, impact and process evaluations of complex WASH trials, formative research in low-and middle-income countries, and humanitarian contexts. She has worked on projects linked to behaviour change for menstrual health, handwashing with soap and preventative behaviour linked to Covid19. She has worked in multiple countries with multiple implementing partners in East and West Africa.

Hugh Sharma Waddington is Assistant Professor at LSHTM and London International Development Centre. He specialises in impact evaluation and evidence synthesis on global development and environmental health topics. He is Co-Chair of the Campbell Collaboration Climate Solutions Coordinating Group, and a Methods Editor with Campbell Systematic Reviews and the Journal of Development Effectiveness. He founded the Systematic Reviews Programme and London Office of the International Initiative for Impact Evaluation (3ie), and has been employed in policy and policy research positions in the Government of Rwanda, the World Bank, the UK National Audit Office and the University of Sussex.

Kaposo Mwambuli has nearly two decades of experience in leading successful projects across Africa, specialising in marketing, consumer insights, and strategic communication. He previously spent more than a decade executing development programmes in Tanzania with the World Bank, before joining WaterAid’s Dar es Salaam office where he served as Director of Policy, Research and Sector Engagement. Today, Kaposo is the Country Team Leader for Project CLEAR in Tanzania, working in close partnership with the Government of Tanzania to deliver behaviour change at scale. He is also the Tanzania and Malawi Country Engagement Lead for the Climate Resilient Infrastructure Development Facility (CRIDF) which manages transboundary water resource management issues for the Southern African Development Community. Kaposo has an MBA in Marketing, Innovation and Change Management from the Maastricht School of Management in the Netherlands.

References

Ajzen

(1991). The theory of planned behavior. Organizational Behavior and Human Decision Processes, 50(2), 179–211. https://doi.org/10.1016/0749-5978(91)90020-t

Aunger

(2020). Reset: An introduction to behaviour centred design. Oxford University Press.

Aunger

Curtis

(2016). Behaviour centred design: Towards an applied science of behaviour change. Health Psychology Review, 10(4), 425–446. https://doi.org/10.1080/17437199.2016.1219673

Bandura

(1986). Social foundations of thought and action: A social cognitive theory. Prentice Hall.

Barker

R. G.

(1968). Ecological Psychology: Concepts and methods for studying the environment of human behavior. Stanford University Press.

Darmawan

I. K.

(2020). Unbranded vs. branded direct-to-consumer advertising (dtca) using social media influencers: Examining the effects of message type and disclosure. University of Minnesota. https://conservancy.umn.edu/bitstream/handle/11299/214985/Darmawan_umn_0130M_21229.pdf?sequence=1

Green

K. M.

Crawford

B. A.

Williamson

K. A.

DeWan

A. A.

(2019). A meta-analysis of social marketing campaigns to improve global conservation outcomes. Social Marketing Quarterly, 25(1), 69–87. https://doi.org/10.1177/1524500418824258

Lefebvre

R. C.

(2000). Theories and models in social marketing. In Bloom

P. N.

Gundlach

P. N.

(Eds.), Handbook of marketing and society. Sage Publications.

Savović

Jones

Altman

Harris

Jűni

Pildal

Als-Nielsen

Balk

Gluud

Ioannidis

Schulz

Beynon

Welton

Wood

Moher

Deeks

Sterne

(2012). Influence of reported study design characteristics on intervention effect estimates from randomised controlled trials: Combined analysis of meta-epidemiological studies. Health Technology Assessment, 16(35), 1–82. https://doi.org/10.3310/hta16350

10.

Strecher

V. J.

Rosenstock

I. M.

(1997). The health belief model. In Baum

(Ed.), Cambridge handbook of psychology, health and medicine (pp. 113–117). Cambridge University Press.

11.

Tweneboah-Koduah

E. Y.

Coffie

I. S.

(2022). Social distancing, hand washing and handshaking behaviour during and beyond coronavirus pandemic: A social marketing perspective. Social Marketing Quarterly, 28(4), 288–306. https://doi.org/10.1177/15245004221134208

12.

Wolff

(2021). COVID-19 vaccination intentions: The theory of planned behavior, optimistic bias, and anticipated regret. Frontiers in Psychology, 12(2021), 648289. https://doi.org/10.3389/fpsyg.2021.648289

13.

Wollast

Schmitz

Bigot

Luminet

(2021). The theory of planned behavior during the COVID-19 pandemic: A comparison of health behaviors between Belgian and French residents. PloS One, 16(11), e0258320. https://doi.org/10.1371/journal.pone.0258320

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.50 MB