18.2 Marketing
18.2.1 Trust
Measuring trust algorithmically using social media data (Roy et al. 2017)
18.2.2 Sentiment
(J. Hartmann et al. 2023) Accuracy and Application of Sentiment Analysis
Sentiment is core to human communication.
Marketing uses sentiment analysis for:
Social media.
News articles.
Customer feedback.
Corporate communication.
Available sentiment analysis methods:
Lexicons: link words/expressions to sentiment scores.
Machine learning: complex but potentially more accurate.
Study introduces an empirical framework to:
- Evaluate method suitability based on research questions, data, and resources.
Meta-analysis conducted on:
272 datasets.
12 million sentiment-labeled documents.
Findings:
Transfer learning models top performance.
These models may not always meet leaderboard benchmarks.
Transfer learning models are, on average, >20% more accurate than lexicons.
Performance influenced by:
Number of sentiment classes.
Text length.
Study offers:
SiEBERT - a pre-trained sentiment analysis model.
Open-source scripts for easy application.
18.2.3 Purchase Intention
(J. Hartmann et al. 2021) The Power of Brand Selfies
RoBERTa-based model: https://huggingface.co/j-hartmann/purchase-intention-english-roberta-large
Smartphones simplify sharing branded imagery.
Study categorizes social media brand imagery.
Identified image types:
Packshots (just the product).
Consumer selfies (consumer’s face with brand).
Brand selfies (product held, no visible consumer).
Convolutional neural networks used to recognize image types.
Language models analyze social media responses to 250,000+ brand-image posts from 185 brands on Twitter & Instagram.
Findings:
Consumer selfies lead to more likes and comments.
Brand selfies induce higher purchase intentions.
Traditional social media metrics may not fully capture brand engagement.
Display ad results:
- Higher click-through rates for brand selfies than consumer selfies.
Lab experiment indicates self-reference affects image responses.
Machine learning can decipher marketing insights from multimedia content.
Image perspective impacts actual brand engagement.
18.2.4 Brand Reputation
- Measuring brand reputation from Twitter data (Roland T. Rust et al. 2021)
18.2.5 Capabilities
(Shantanu Dutta, Narasimhan, and Rajiv 1999) Marketing capability is critical for high-technology markets
Insights
A company’s sales are increased by a solid foundation of innovative technologies that positively affect consumers’ perceptions of the benefits of its product’s externalities.
For businesses with a strong technological foundation, marketing capability has the biggest impact on the production of innovation that has been quality-adjusted. The companies that stand to benefit the most from great marketing capabilities are those with a solid R&D foundation.
- because a company may have tremendous R&D capabilities but be unable to translate them into commercially viable items due to weak marketing skills
The interaction of marketing and R&D capabilities is the most significant factor in determining a firm’s performance.
- High tech firm need to have both the ability to come up with innovation constantly and the ability to commercialize these innovations.
A firm’s capability is defined as “its ability to deploy the resources (inputs) available to it to achieve the desired objective(s) (output).”
Logically, the higher the functional capability a firm has, the more efficiently it is able to deploy its productive inputs to achieve its functional objectives.
Equivalently, the lower the functional inefficiency, the higher the functional capability of the firm
Operationalization
Marketing Capability
Sales = f(technological base, advertising stock, stock of marketing expenditure, investment in customer relationships, installed base)
Weight for marketing expenditure = 0.5 (Shantanu Dutta, Narasimhan, and Rajiv 2005, 281)
Advertising stock weight = 0.4 (Peles 1971) or 0.5 (Zhan Wang and Kim 2017)
R&D Capability
2 dimensions of the quality of technological output
innovativeness (Trajtenberg 1990b, 1990a)
width of applicability (Jaffe, Trajtenberg, and Henderson 1993)
Quality-adjusted technological output = f(technological base, cumulative R&D expenditure, marketing capability)
R&D expenditure weight = 0.4 (Shantanu Dutta, Narasimhan, and Rajiv 2005, 281)
Operations Capability
Cost of production = f(output, cost of capital, labor cost, technological base, marketing capability)
Variables
| Label | Variable |
|---|---|
| Sales | |
| Did not use raw patent count because quality matters. | |
Innovative-adjusted technological output The number of times the patents of a firm have been cited (citation-weighted patent count) |
Replication
As the time of this writing, the US Patents by WRDS dataset only covers from 2011 to 2019. Hence, I can only replicate the results in this period but not the years in the study (1985 - 1994). Alternatively, you can visit USPTO to download the raw dataset and create your own matching algorithm based on company names.
wrdsapps_patents_link is from US Patents by WRDS / Compustat Link
wrdsapp_patents is from US Patents by WRDS / Patents
- forward citations counts as of Dec 31st 2019
wrdsapp_patents_citations is from US Patents by WRDS / Citations
library(tidyverse)
library(lubridate)
totalq <- read.csv(file.path("data/totalq.csv.gz"))
capability <- read.csv(file.path("data/capability.csv.gz")) %>%
# fix the length of numeric variables
mutate(gvkey = str_pad(as.character(gvkey),width = 6, pad = "0", side = "left"),
sic = str_pad(as.character(sic), width = 4, pad = "0", side = "left")) %>%
# need to have data on advertising, r&d, sga, cogs
filter(!is.na(xad) & !is.na(xrd) & !is.na(xsga) & !is.na(cogs))
wrdsapps_patents_link <-
read.csv("data/patents/wrdsapps_patents_link.csv.gz") %>%
mutate(gvkey = str_pad(
as.character(gvkey),
width = 6,
pad = "0",
side = "left"
)) %>%
# get industry data
inner_join(capability %>% select(gvkey, sic) %>% distinct(), by = "gvkey")
# view(head(wrdsapps_patents_link))
wrdsapps_patents <- read.csv("data/patents/wrdsapps_patents.gz")
# view(head(wrdsapps_patents))
wrdsapps_patents_citations <- read.csv("data/patents/wrdsapps_patents_citations.csv.gz") %>%
# correct dates
mutate(grantdate = as.Date(as.character(grantdate), "%Y%m%d"),
cited_pat_gdate = as.Date(as.character(cited_pat_gdate), "%Y%m%d")) %>%
# get year
# grant year here is the year that the new patent that cite the old one was granted.
mutate(grantyear = year(grantdate),
cited_pat_gyear = year(cited_pat_gdate)) %>%
# get only those patents belong to firms in the Compustat database
left_join(
wrdsapps_patents_link %>%
select(patnum, sic) %>%
rename(cited_patnum_sic = sic),
by = c("cited_patnum" = "patnum")
) %>%
left_join(
wrdsapps_patents_link %>%
select(patnum, sic) %>%
rename(patnum_sic = sic),
by = c("patnum" = "patnum")
)
# view(head(wrdsapps_patents_citations, 100))
wrdsapps_patents_citations_by_year <- wrdsapps_patents_citations %>%
select(patnum, grantyear, cited_patnum) %>%
count(cited_patnum, grantyear) %>%
rename(citation_count = n)
# check
# wrdsapps_patents_citations_by_year %>%
# filter(cited_patnum == "01044494") %>%
# view()
# wrdsapps_patents_citations %>%
# filter(cited_patnum == "01044494") %>%
# view()Innovativeness-adjusted technological output:
- Calculate the average number of citations receive by all patents belonging the firms in the sample within one industry (defined by sic1, sic2, sic3, sic4)
- The weight assigned to a firm’s patents = # of citations / industry sample average
- Tech_innv = sum of citation-weighted patents within a firm in a year.
# calculate industry average citation based on
# sic1
tech_innv1 <- aggregate(x = tech_innv$citation_count,
by = list(tech_innv$sic1),
FUN = mean) %>%
rename(sic1 = 1,
mean_cite_ind1 = 2)
# sic2
tech_innv2 <- aggregate(x = tech_innv$citation_count,
by = list(tech_innv$sic2),
FUN = mean) %>%
rename(sic2 = 1,
mean_cite_ind2 = 2)
# sic3
tech_innv3 <- aggregate(x = tech_innv$citation_count,
by = list(tech_innv$sic3),
FUN = mean) %>%
rename(sic3 = 1,
mean_cite_ind3 = 2)
# sic4
tech_innv4 <- aggregate(x = tech_innv$citation_count,
by = list(tech_innv$sic),
FUN = mean) %>%
rename(sic = 1,
mean_cite_ind4 = 2)
# merge all four sic types together
tech_innv_sic <- tech_innv4 %>%
mutate(
sic3 = substr(sic, 1, 3),
sic2 = substr(sic, 1, 2),
sic1 = substr(sic, 1, 1)
) %>%
full_join(tech_innv3, by = "sic3") %>%
full_join(tech_innv2, by = "sic2") %>%
full_join(tech_innv1, by = "sic1") %>%
select(-c(sic1, sic2, sic3))
# note that a modification from Dutta's paper is that I use the average number of citations received by all the patents belonging the firms in the sample within one industry (because industry average citations might differ)
# Dutta's paper did not do this because they consider only one industry
# calculate the Tech_innv
tech_innv <- wrdsapps_patents_link %>%
select(gvkey, patnum, sic) %>%
inner_join(wrdsapps_patents_citations_by_year, by = c("patnum" = "cited_patnum")) %>%
# match with the firm and year data
filter(gvkey %in% capability$gvkey) %>%
inner_join(tech_innv_sic, by = "sic") %>%
# create the weight assigned to a firm's patent
mutate(weight1 = citation_count / mean_cite_ind1,
weight2 = citation_count / mean_cite_ind2,
weight3 = citation_count / mean_cite_ind3,
weight4 = citation_count / mean_cite_ind4) %>%
group_by(gvkey, grantyear) %>%
# create the citation-weighted patent count
summarize(tech_innv1 = sum(weight1),
tech_innv2 = sum(weight2),
tech_innv3 = sum(weight3),
tech_innv4 = sum(weight4)) %>%
ungroup()
rm(tech_innv_sic,
tech_innv1,
tech_innv2,
tech_innv3,
tech_innv4)Width-of-applicability-technological output 1. Calculate the proportion of citations received by a patent from firms belonging outside of focal SIC code (sic1, sic2, sic3, sic4) = # of citations received by a patent from firms outside the focal SIC code / the total number of citation the patent received. 2. The weight assigned to a firm’s patent = the proportion of outside citations for the patent/ the industry sample average proportion 3. Tech_width = The sum of the “proportion-of-outside citation”-weighted patent in a year for a firm
wrdsapps_patents_citations_out_ind <- wrdsapps_patents_citations %>%
# select patent number with data on industry
filter(!is.na(cited_patnum_sic) & !is.na(patnum_sic)) %>%
# create variables indicating whether the new and old patents belong in the same industry
mutate(sim4 = if_else(patnum_sic == cited_patnum_sic, 1, 0),
sim3 = if_else(substr(patnum_sic,0,3) == substr(cited_patnum_sic,0,3), 1, 0),
sim2 = if_else(substr(patnum_sic,0,2) == substr(cited_patnum_sic,0,2), 1, 0),
sim1 = if_else(substr(patnum_sic,0,1) == substr(cited_patnum_sic,0,1), 1, 0)) %>%
select(
patnum,
cited_patnum,
grantyear,
cited_pat_gyear,
cited_patnum_sic,
patnum_sic,
contains("sim")
)
total_citation <- wrdsapps_patents_citations_out_ind %>%
group_by(cited_patnum, cited_pat_gyear) %>%
count()
# view(head(total_citation))
# get per patent, per year, the number of outside-industry citations
# sic1
outside_citation_sic1 <- wrdsapps_patents_citations_out_ind %>%
group_by(cited_patnum, cited_pat_gyear, sim1) %>%
count() %>%
ungroup() %>%
# get only citations outside of the patent's industry
filter(sim1 == 0) %>%
select(-c(sim1)) %>%
rename(n_sim1 = n)
# sic2
outside_citation_sic2 <- wrdsapps_patents_citations_out_ind %>%
group_by(cited_patnum, cited_pat_gyear, sim2) %>%
count() %>%
ungroup() %>%
# get only citations outside of the patent's industry
filter(sim2 == 0) %>%
select(-c(sim2)) %>%
rename(n_sim2 = n)
# sic3
outside_citation_sic3 <- wrdsapps_patents_citations_out_ind %>%
group_by(cited_patnum, cited_pat_gyear, sim3) %>%
count() %>%
ungroup() %>%
# get only citations outside of the patent's industry
filter(sim3 == 0) %>%
select(-c(sim3)) %>%
rename(n_sim3 = n)
# sic4
outside_citation_sic4 <- wrdsapps_patents_citations_out_ind %>%
group_by(cited_patnum, cited_pat_gyear, sim4) %>%
count() %>%
ungroup() %>%
# get only citations outside of the patent's industry
filter(sim4 == 0) %>%
select(-c(sim4)) %>%
rename(n_sim4 = n)
tech_width <- total_citation %>%
full_join(outside_citation_sic1, by = c("cited_patnum", "cited_pat_gyear")) %>%
full_join(outside_citation_sic2, by = c("cited_patnum", "cited_pat_gyear")) %>%
full_join(outside_citation_sic3, by = c("cited_patnum", "cited_pat_gyear")) %>%
full_join(outside_citation_sic4, by = c("cited_patnum", "cited_pat_gyear")) %>%
# fill in 0 (those patents without any citations outside of their industry)
replace(is.na(.), 0) %>%
mutate(weight1 = n_sim1 / n,
weight2 = n_sim2 / n,
weight3 = n_sim3 / n,
weight4 = n_sim4 / n) %>%
# get company gvkey for each patent
inner_join(wrdsapps_patents_link %>%
select(gvkey, patnum, sic),
by = c("cited_patnum" = "patnum")) %>%
# step 3: get proportion-of-outside citation-weighted patent
group_by(gvkey, cited_pat_gyear) %>%
summarize(tech_width1 = sum(weight1),
tech_width2 = sum(weight2),
tech_width3 = sum(weight3),
tech_width4 = sum(weight4)) %>%
ungroup()
rm(outside_citation_sic1,
outside_citation_sic2,
outside_citation_sic3,
outside_citation_sic4,
total_citation)
rm(wrdsapps_patents,
wrdsapps_patents_citations,
wrdsapps_patents_citations_by_year,
wrdsapps_patents_citations_out_ind,
wrdsapps_patents_link)18.2.5.1 Marketing Capabilities
Bahadir, Bharadwaj, and Srivastava (2008)
Marketing Capability: (Xiong and Bharadwaj 2013)
Wiles, Morgan, and Rego (2012)
Dinner, Kushwaha, and Steenkamp (2018)
Marketing Alliance Capabilities: Swaminathan and Moorman (2009)
Digitized Selling Capabilities: (D. S. Johnson 2005)
Big Data Capabilities: (J. S. Johnson, Friend, and Lee 2017)
Social CRM Capabilities: (Trainor et al. 2014)
Customer relationship capabilities: (Zhan Wang and Kim 2017)
library(frontier)
mar_cap <-
sfa(
log(sales) ~ sub_market,
log(adstock) + log(marketingstock) + log(techbase) + log(receivable) + log(installedbase),
data = capability,
ineffDecrease = T, # inefficiency decreases the endogenous variable for estimating a production function
timeEffect = T # Error Compoennts Frontier are time-variant
)
efficiencies(mar_cap)# defunct
# install.packages("FEAR")
# install.packages("sfaR")
library(sfaR)
mar_cap <-
sfacross(
log(sales) ~ sub_market,
log(adstock) + log(marketingstock) + log(techbase) + log(receivable) + log(installedbase),
data = capability
# ineffDecrease = T, # inefficiency decreases the endogenous variable for estimating a production function
# timeEffect = T # Error Compoennts Frontier are time-variant
)
efficiencies(mar_cap)(Neil A. Morgan, Slotegraaf, and Vorhies 2009) Linking marketing capabilities with profit growth
Objective: Examine the connection between a firm’s marketing capabilities and its profit growth, with a particular emphasis on how specific marketing functions impact the composite components of profit growth.
Context: Although profit growth stands as a pivotal determinant of a firm’s stock price, there’s a limited understanding of how integral marketing capabilities interlink with this growth trajectory.
Methodology:
Engaged a cross-industry dataset from 114 firms.
Delineated the exploration into three cornerstone marketing capabilities: market sensing, brand management, and customer relationship management (CRM).
These capabilities were then juxtaposed with the dual facets of profit growth: revenue growth and margin growth.
Key Findings:
The scrutinized marketing capabilities exhibited both direct and synergistic influences on the growth rates of revenue and margin.
A pivotal revelation was the counteractive effects of brand management and CRM capabilities on the growth rates of revenue and margin. Specifically, while one capability might promote revenue growth, it might simultaneously inhibit margin growth (and vice versa).
Such intricate dynamics imply that a surface-level analysis, overlooking the nuanced contributions to revenue and margin growth, could obscure the genuine relationships tying marketing capabilities to the overarching profit growth trajectory.
18.2.5.4 Organization Capabilities
(Grewal and Slotegraaf 2007) Embeddedness of Organizational Capabilities
Core Issue:
- Managers need to efficiently use limited resources to build lasting organizational capabilities for a sustainable competitive edge.
Challenge:
- Neglecting the intricate underlying processes of organizational capabilities can hinder understanding their impact on competitive advantage.
Key Insight:
- Managerial decisions regarding resource usage directly impact the depth at which capabilities are ingrained in the organization, termed “capability embeddedness.”
Research Method:
Introduced a hierarchical composed error structure framework.
Uses cross-sectional data and can be applied to panel data.
Case Study: Retailing
Organizational capability embeddedness directly influences retailer performance.
This influence remains significant even when considering both tangible and intangible resources/capabilities.
Takeaways:
Acknowledging how resources and capabilities affect performance at various organizational layers helps managers make informed decisions.
Essential to recognize whether the objectives of capabilities align (convergent) or differ (divergent).
The alignment or misalignment of these objectives can dictate how much embedded capabilities boost firm performance.
18.2.5.3 Social Media Strategics Capabilities
(B. Nguyen et al. 2015)