The Impact of Product Variety on Quality Conformance in Continuous Process Manufacturing: A Quantitative Investigation in a Chemical Industry Context

Abstract

Product variety in manufacturing has increased significantly, driven by technological advancements and growing demand for customisation. To meet diverse customer preferences, companies often overextend their portfolios without fully considering the resulting impact on manufacturing effectiveness. This study investigates how product variety is associated with quality conformance in continuous process manufacturing, an area underexplored in the existing literature, which predominantly focuses on discrete or assembly-based operations. Utilising production data from a large chemical manufacturer, logistic regression analysis was applied to examine how variety-related engineering parameters relate to the probability of non-conformance at the big-bag level. The analysis shows that ramp-ups, especially those associated with major changeovers, and short production uptimes are correlated with an increased likelihood of quality issues. Very infrequent production also appears to increase quality risks. Contrary to learning-curve expectations, products with medium and low production intensity showed lower odds of non-conformance than high-intensity products. These findings indicate the implications of product variety in continuous manufacturing environments. By identifying variety parameters that appear to contribute to quality risks, this study offers initial guidance for production planners and product portfolio managers aiming to balance product variety with quality conformance and overall manufacturing effectiveness.

1. Introduction

Manufacturing companies have significantly expanded their portfolios in response to increasing demands for customised solutions, product digitalisation, and competitor globalisation [1]. While new product introductions are essential for driving repeat purchases, companies risk overpopulating their portfolios if these introductions are not accompanied by active efforts to reduce product variety [2]. Increasing product variety introduces several engineering and performance challenges for manufacturers [3], and an assortment with high variety can ultimately only create value for customers if the product variants can be manufactured without incurring excessive penalties in price, quality, or delivery performance [4].

One of the most critical performance challenges associated with increasing product variety is its impact on quality conformance, a relationship widely cited in the literature [5,6,7,8]. An increase in product variety might be expected to cause a decrease in quality in a manufacturing environment in multiple ways. As variety increases, the production volume per unit is reduced, making it more difficult to perform statistical process control and monitor process behaviour, thus affecting the end quality [9]. Similarly, as the per-unit product volume is diffused, quality could suffer, as defects tend to decline with longer production runs. A higher variety with a lower volume per part also increases the risk of production line stoppage due to stockouts, which can cause quality problems [10]. Moreover, increased product variety leads to greater operator option complexity, leaving production workers with a more complicated and less predictable array of parts to assemble [11].

While the literature broadly acknowledges the adverse relationship between product variety and quality conformance, empirical evidence remains limited. Moreover, most existing studies have focused on the automotive industry. For instance, ref. [11] analysed data from an international motor vehicle programme covering the period 1985–1990, while [10] investigated the impact of product variety using data from a General Motors assembly plant. Ref. [12] conducted a case study in two Volvo Car Corporation assembly plants. Similarly, ref. [13] examined three Western European automotive manufacturers representing heavy trucks, small cars, and premium cars. Finally, ref. [14] integrated data from multiple automotive-specific repositories to examine the relationship between product variety and recall rates. Beyond the automotive sector, only a single empirical study exists in the textile industry [15]. Consequently, most existing empirical research has been confined to automotive assembly operations, leaving a critical gap in understanding how product variety affects quality in other industry contexts, such as continuous process manufacturing. In fact, some authors have specifically pointed to the lack of academic inquiry into product variety and complexity in the process manufacturing industry [16,17]. As a result, the current literature provides little guidance for process manufacturing companies on which aspects of product variety affect quality conformance, leaving companies aiming to improve quality with a single way forward, namely, to reduce variety. This lack of specificity is unhelpful for both companies looking to increase their variety and those willing to reduce it, as there are few indications in the literature concerning which variety to eliminate. Accordingly, recommendations in this area require further expansion and development to become actionable insights.

The present article aims to outline the problems of managing quality conformance in environments characterised by high product variety and continuous process manufacturing. In this context, quality conformance is defined as ‘the degree to which a specific product conforms to a specification’, following the manufacturing-based definition proposed by [18] (p. 26). More specifically, the article will identify a way to analyse quality conformance data, outline a suitable unit of analysis, and explore the mechanisms through which product variety affects quality conformance. Notably, the chosen context, namely continuous process manufacturing, is a context rarely investigated in the literature on product variety, and consequently, the research approach will have an exploratory intention. Lastly, the paper aims to propose actionable insights for engineering and supply chain managers concerning which elements of product variety to reduce and which products to eliminate to reduce off-quality. Based on these objectives, the present study will develop a model that identifies specific product variety aspects of continuous process manufacturing as variables affecting quality conformance. By testing this model (which in the end corresponds to the adverse relationship between product variety and quality conformance taken implicitly for granted by the literature in every context, even though not thoroughly operationalised and assessed in continuous manufacturing), we are able to answer the following fundamental research question pursued by the present paper:

Research question (RQ1): Which product variety-related factors are most likely to influence product quality conformance in continuous process manufacturing?

To pursue the stated objectives, this study adopts a quantitative modelling approach. Quantitative modelling is common for research surrounding product variety and manufacturing performance (see, for example, [19,20,21]) as it enables very reliable insights based on strong empirical evidence. For the present study, the use of data from a continuous process manufacturing setting allows for a systematic analysis of large volumes of observations, which would be difficult to interpret through qualitative approaches alone. Specifically, logistic regression is employed to model the likelihood of quality conformance issues as a function of product variety-related engineering parameters. The article is structured in seven sections. First, the literature concerning the impact of product variety on business process performance and quality conformance is reviewed. Next, the methodology is outlined, including the justification for the case context, variable selection, and modelling approach. This section is followed by the presentation of results, a discussion of the findings, concluding remarks, and directions for future research.

2. Theoretical Background

The impact of product variety on companies’ sales and operational performance has long been a topic of interest in both marketing and operations research [6]. Product variety has attracted marketers’ interest due to its assumed value for customers, with literature pointing to the benefits of product variety in satisfying heterogeneous needs [22]. Wider product assortments help companies to meet diverse demands and enable customers to make purchases that better align with their individual needs [23]. Customers also perceive companies that offer a wide variety of products as having greater category commitment and expertise [24]. Collectively, these findings imply a positive relationship between product variety and sales growth [25].

However, while an increase in product variety may positively influence sales and market share, it can also lead to negative implications for business performance [3]. Extensive product variety is likely to have implications across the entire value chain. First, research shows that perceived variety and purchase likelihood follow an inverted U-shape, meaning that sales rise with product variety up to a certain point, after which further variety reduces sales [26]. Several factors begin to counter the positive relationship between product variety and customer demand once the portfolio size exceeds manageable levels. Increasing product variety may, for example, reduce logistic service quality [22], increase customer confusion [4], and increase requirements for sales force training [27]. Extending the portfolio may also imply product cannibalisation, where the demand for one product comes at the expense of others offered by the same company [28].

Second, product variety is likely to reduce demand predictability [29], which in turn may introduce forecast bias into the system, meaning the tendency to over- or underestimate demand. This relationship arises not only from the direct proliferation of the product portfolio but also from increased interaction complexity, such as product substitution and cannibalisation [30].

Next, as the number of products in the portfolio increases, manufacturing complexity tends to rise [31]. Greater product variety disrupts manufacturing flow by reducing batch sizes and increasing the number of setups [27]. When variety grows, production lines must handle a wider range of products, often resulting in smaller production orders and shorter run times for each order. This, in turn, increases the frequency of changeovers, which can destabilise process settings and cause unfavourable shifts in production speed between consecutive batches. Moreover, extensive product variety complicates order sequencing and production scheduling [19]. As idle time on multi-purpose manufacturing lines is consumed with the introduction of additional variety, production tends to be either in continuous operation or in setup mode, leading to exponential increases in both inventory levels and cash-to-cash cycle times [28]. Together, these effects highlight how rising product variety places significant strain on manufacturing performance, impacting efficiency, throughput, and overall operational stability. Supporting this view, ref. [17] conducted a systematic literature review on product variety, product complexity, and manufacturing operational performance, and found a consistently negative relationship between product variety and manufacturing operational performance. However, studies that examine this relationship are primarily confined to the automotive and electronics sectors. This sectoral concentration is likely explained by the well-documented progression in these industries from mass production to the early adaptation of mass customisation [11], making them natural case studies when the concept of balancing variety and operational efficiency first emerged.

Several researchers have also hypothesised an inverse relationship between product variety and quality conformance. First, ref. [9] argued that increasing product variety on an automotive assembly line could reduce operator line balance, decrease labour efficiency, and necessitate costly rework. The authors also reasoned that quality could suffer if variety is added, as it is harder to perform statistical process control on parts with limited production history [9]. Ref. [11] found no statistically significant correlation between the number of product variants produced at an automotive plant and assembly defects, although determining this correlation was not the principal focus of their study. Next, in an empirical study of an automotive assembly plant, ref. [10] found that major rework is positively related to option variability, that is, a product variety measure indicating the standard deviation in the number of key options per car. More specifically, the authors provided statistical indications that reducing option variability by 10% from its mean value would decrease major assembly rework by 8.3%. Similarly, ref. [12] showed a significant positive correlation between operator choice complexity and assembly errors, using 16 weeks of data from seven automotive pre-assembly lines. The complexity measure was derived from the demand and number of product variants at each station, while the assembly deficiencies were derived from 11 subcategories with data extracted from the internal quality system. Ref. [13] found that increasing product variety challenges the effectiveness of traditional quality management systems. Techniques such as statistical process control and Six Sigma typically rely on large batches of similar products to identify and eliminate variation. However, greater product variety reduces batch sizes and the likelihood of repetitive processes, making it difficult to collect sufficient data to apply these methods effectively. Furthermore, product variety challenges the classic plan–do–check–act continuous improvement cycle, as processes become less stable and predictable. Next, ref. [14] investigated the effect of product variety on manufacturing-related product recalls (quality) using data on automotive recall occurrences from 2000 to 2006. Product recalls, measured as the total number of manufacturing recalls for a given car model, were defined as the dependent variable, while product variety, plant variety, and manufacturing line utilisation served as independent variables. Following an approach similar to [10,11], factory-installed options such as air conditioning and electronic suspension were used to measure product variety, whereas plant variety was measured as the ratio of the number of models built per assembly line. The findings revealed a positive relationship between product variety and manufacturing recalls, whereas plant variety had no significant effect. In addition to the research focusing on automotive assembly, ref. [15] examined the effect of product mix (unique warp and fill thread combinations) on quality conformance using four years of data from three textile manufacturing plants. She found that increases in major setups are associated with a reduction in quality performance. Moreover, her findings revealed that two distinct aspects of product mix composition, fabric weight and warp beam construction, are negatively correlated with quality conformance. Lastly, ref. [32] provided empirical evidence to support the assertion that higher product variety and inventory levels increase defect rates in a retail setting.

The literature shows that, despite wide agreement among researchers on the adverse impact of product variety on quality conformance, the mechanisms and extent of this impact have received limited investigation. Furthermore, since the indications are mainly confined to automotive assembly operations, the investigation should be expanded to other manufacturing contexts where quality issues may have significant economic impacts. Continuous process manufacturing offers a particularly interesting context for such an investigation and has distinct differences from automotive assembly. Unlike the assembly process, where the different parts are pieced together, continuous process manufacturing is a transformation process in which the products’ physical and/or chemical characteristics change. Consequently, product variety in continuous process manufacturing differs inherently from that of discrete manufacturing. While assembly operations are labour-intensive, continuous process manufacturing is a technology-intensive environment with a high presence of automation, meaning that nonconformities will not only result in manpower loss but will also have implications for machine availability. Continuous process manufacturing is typically considered an area with high repetitiveness and low product changeover frequency [33]. However, as with the assembly process, in several cases, this type of manufacturing may have to accommodate smaller batch sizes and considerable amounts of variety.

3. Methods

3.1. Choice of the Specific Case

Data were obtained from a Scandinavian continuous process manufacturing company operating within the chemical industry to test the association between product variety and quality conformance and explore which variety-related factors have the strongest influence on product quality conformance in a continuous process manufacturing context. Focusing on a single company allowed us to obtain the high-quality data essential for the kind of exploration pursued and enabled us to rule out several factors that could have affected the analyses in a cross-company study.

The analysis was scoped to one continuous multi-purpose manufacturing line from which a diverse range of products is produced, allowing for the analysis of the effects of product variety on quality conformance.

3.2. Data Collection

Data from 25 consecutive months of production (1 January 2022 to 31 January 2024) were obtained from the case company, capturing production records for 34,875 packed big bags across 56 different product variants. This extended observation period ensures that production events were captured under a wide range of operational conditions, including varying product mixes, production schedules, and order sizes. Notably, the considered production was not affected by systematic seasonality or long-term trends that would necessitate a longer observation period. Consequently, the period of observation can be considered representative of the production under examination.

The data were acquired from the company’s application programming interface and the enterprise resource planning system and subsequently compiled into a single dataset. Most production data originate from sensor recordings on the production line, which are automatically stored via the application programming interface. Additional order-level information, such as production scheduling parameters and quality outcomes, is logged by trained company professionals in the enterprise resource planning system, following established protocols. Both systems underpin multiple critical business functions, including quality control, production planning, costing, and overall equipment effectiveness monitoring, and are therefore subject to continuous oversight and validation. Because these data streams support key operational and regulatory processes, the measurement systems can be considered reliable for the present analysis. The investigated multi-purpose manufacturing line did not facilitate the production of research and development batches, making the collected data representative of full-scale manufacturing. A small number of records with incomplete or inconsistent information (estimated to be below 1% of total production data) were excluded during data preparation to ensure accuracy and comparability. The final dataset includes information on time period, volume, order number, product variant, and product family and a quality indicator for each big bag produced.

3.3. Unit of Analysis

The basic unit of analysis in this study was a packed big bag: a large, flexible industrial container (typically made of woven polypropylene) filled with dry bulk materials such as powders, granules, or fertilisers. Each big bag represents a single, self-contained unit used for the storage, handling, and transport of the material. The big bags are loaded directly from the production line as part of a continuous manufacturing process.

Quality samples are taken at regular intervals just before packaging, but evaluating these takes time. If a quality sample is later found to fall outside specification, the entire corresponding big bag is rejected. Consequently, the packaged big bags constitute the smallest quality-relevant unit in the manufacturing process.

The choice of the big bag as the unit of analysis allowed for a significantly larger dataset than would have been possible had the analysis been conducted at, for example, the production order level. In continuous manufacturing, a single production order may run for several days, during which a variety of factors can influence quality conformance. Consequently, analysing the effect of the different influencing factors at the order level would be inappropriate. This increased granularity enhanced the statistical power of the analysis and allowed for more robust insights into quality performance.

3.4. The Dependent Variable

Product quality was selected as the dependent variable in accordance with the focused relationship between product variety and quality conformance. A value of 1 indicates the occurrence of quality nonconformance, while a value of 0 indicates no quality issues. Nonconformance was identified based on the company’s internal quality control metrics, which classify a product big bag as off-quality if it fails to meet established tolerance levels for indicators such as dimensions, colour, and chemical properties.

This operationalisation of quality conformance as a binary outcome is consistent with the manufacturing-based view of quality as conformance to specification [18] and with operational practice in continuous process industries, where each output unit is either accepted, reworked, or rejected. In the investigated setting, samples taken just before packaging can trigger the rejection of the corresponding big bag, which therefore represents the smallest quality-relevant and economically meaningful decision unit. Modelling the probability of nonconformance via logistic regression, therefore, aligns the statistical target with the managerial decision boundary. Moreover, the application of binary outcome measures is common in empirical studies of manufacturing and quality. For example, ref. [34] formulates off-quality detection as a binary classification problem. Similarly, ref. [35] use logistic regression with a company-informed binary output variable to model satisfactory/unsatisfactory production output.

3.5. The Independent Variables—The Starting Point of Identification

Following [15], the analysis was supplemented with a range of engineering parameters that describe the process settings of the 56 different product variants produced during the 25 months. Initial variable selection combined insights from the literature with input from company experts. On one hand, we introduced managers to concepts identified in academic research. On the other hand, we asked them to identify factors they believed or suspected might influence quality conformance in practice. We deliberately chose not to rely solely on a deductive approach because the limited literature on continuous manufacturing increases the risk of overlooking variables that are critical in the specific industrial context. Moreover, involving practitioner insights ensures that the study reflects both established theory and the tacit knowledge embedded in day-to-day operations.

First, production ramp-up was included to capture the operational challenges associated with setups and changeovers between product variants. Company experts highlighted the large number of changeovers resulting from product variety and the associated ramp-up periods, during which products are manufactured under suboptimal conditions. This observation aligns with the findings of [15], who reported that increases in major setups in textile manufacturing are associated with reduced quality performance. The severity of each changeover was assessed using a changeover matrix developed by the production planners, which specifies the expected downtime for each changeover.

Second, stakeholders emphasised the total uptime of a production order as an important determinant of quality conformance. This view is supported by the literature, which suggests that quality defects decrease with longer production runs [10].

Next, based on discussions with company stakeholders, production intensity was included to capture the level of production activity, measured as the total number of production orders per product variant. Stakeholders conjectured that higher production intensity improves process familiarity, thereby enhancing quality performance. This assumption aligns with learning curve theory, which posits that as workers or organisations gain experience producing a product, efficiency improves, and the rate of nonconforming units decreases [36].

Next, the stakeholders pointed to the recency of production, meaning the duration since the product’s last production run, as an important determinant of quality conformance. For products not manufactured for an extended period, they argued that production may require ad hoc recalibration and adjustments before reaching a steady state. This argument aligns with [14], who showed that reduced process repetition undermines the stability required for traditional quality management systems, making it more difficult to maintain accurate process data and apply continuous improvement methods effectively.

Lastly, the company experts hypothesised that within-product-family proliferation impacts quality conformance through increased complexity in the manufacturing environment and reduced opportunities for standardisation. This notion is supported by [31], who state that manufacturing complexity tends to rise with an increase in product variety.

As seen from

Table 1. Overview of the initial variables and their theoretical or empirical foundations.

Variable Input	Empirical Origin	Supporting Source
Production ramp-up	✓	[15]
Uptime of a production order	✓	[10]
Number of production orders per product variant	✓	[36]
Duration since the product’s last production run	✓	[14]
Within-product-family proliferation	✓	Company expert input

Note: ✓ indicates variables derived from prior studies.

, this process resulted in an initial set of variables informed by both academic literature and company perspectives. These variables then underwent a refinement process, described in the next subsection, where the final list of independent variables is also presented.

3.6. Data Analysis Method and the Selection of Independent Variables

In this section, we present the specific approach and analysis employed to test the relationship between product variety and quality conformance in a continuous process manufacturing context. All statistical analyses were performed in R (version 4.2.2) using a custom-developed script. The script follows a structured workflow: (i) importing the pre-processed dataset containing merged production and quality data; (ii) conducting descriptive statistics and correlation analysis; (iii) transforming selected continuous variables into categorical factors and performing variable selection using variance inflation factor (VIF) analysis; (iv) estimating logistic regression models with robustness checks; and (v) exporting results for visualisation and reporting. The script employed the tidyverse package suite for data handling and visualisation, car for VIF diagnostics, caret for data partitioning and model evaluation, pROC for ROC and AUC analysis, and broom for tidying model output.

Since the output quality represents a dichotomy, where product big bags are either within the quality specification or not, multiple logistic regression analysis (LRA) was selected to determine the relationship. LRA is an appropriate method, as it is specifically designed to handle binary outcomes, accommodate non-linear relationships, and identify the independent variables that have the highest odds ratios to affect the dependent variable [37].

The variance inflation factor (VIF) was used to guide initial variable selection to address potential multicollinearity. After iterative backwards elimination of variables with high VIF values, six independent variables remained, as shown in

Table 2. Research variables of this study.

Category	Variable Name		Variable Description
Dependent	Y	Product quality	1 if a product big bag was off-quality as per the tolerance levels established by the focal company; 0 otherwise
Independent	X1	Ramp-up with minor changeover	1 if a product big bag was produced during a ramp-up before a minor changeover with an estimated downtime ≤ 60 min; 0 otherwise
	X2	Ramp-up with major changeover	1 if a product big bag was produced during a ramp-up before a major changeover with an estimated downtime > 60 min; 0 otherwise
	X3	Production uptime	Total uptime of the production order, categorised as short (0–3 days), moderate (3–6 days), and long (>6 days)
	X4	Production intensity	Number of production orders per product variant, categorised as low (1–4), medium (5–7), and high (8–11)
	X5	Production frequency	Number of days since a product from the same product family was last produced, categorised as very rare (>365 days), rare (181–365 days), occasional (91–180 days), frequent (31–90 days), and very frequent (0–30 days)
	X6	Product family proliferation	Number of size and shape variations within the product family

.

To ensure the model assumptions were not violated, all the continuous variables were evaluated for skewness. Variables with substantial skewness were transformed to improve interpretability and model performance. For example, continuous variables such as production intensity naturally correlate with higher counts of quality issues. However, this does not mean that a higher rate of quality defects is likely when the intensity of production orders increases. Rather, these products are more likely to showcase quality defects because of their larger production volumes. Transformation also helps to reduce the influence of extreme values and improve the linear relationship between predictors and the log-odds. Furthermore, categorical coding of continuous variables can improve interpretability for applied audiences and ensure alignment with managerial decision boundaries [38]. The continuous variables were inspected visually prior to categorisation to identify natural distributional breaks, and category boundaries were chosen to reflect both these natural breakpoints and operationally meaningful thresholds used within the company. Odds ratios derived from categorical contrasts (e.g., short vs. long run lengths) are directly meaningful for practitioners, thereby enhancing the model’s explanatory value. In the investigated case, the company commonly employs categorical terminology to distinguish between, for example, short, moderate, and long production runs when discussing campaign feasibility, resource requirements, and changeover intensity. Therefore, the adopted categorical transformation aligns with the terminology and decision logic used across the organisation, making the model representative of operational reality.

Lastly, the remaining variables were evaluated for statistical significance using Wald tests. Apart from X6, all the remaining decision variables were found statistically significant. Consequently, X6 was excluded from the model.

3.7. The Final Model Assessed

The final model was estimated using LRA on a dataset comprising 34,875 packed big bags representing 56 different chemical product variants produced over a period of more than two years on the same continuously operating production line. The dependent variable Y represents the binary quality outcome for each packed big bag. Five independent variables, X1 to X5, were included in the model. The model estimates the probability of nonconformance as a function of these variables according to the following expression, where a denotes the intercept and β_i the estimated coefficients for each independent variable. This specification thus quantifies how operational conditions influence the likelihood of quality nonconformance.

logit (P(Y = 1)) = a + β₁ × X1 + β₂ × X2 + β₃ × X3 + β₄ × X4 +β₅ × X5

4. Results

The following section outlines the results of the analysis of the final model. First, the analysis output is presented, along with a description of its structure and an explanation of how to interpret the content. A detailed examination of the relationship between quality conformance and the product variety-related decision variables follows.

4.1. The Overall Model Results

The logistic regression model achieved an overall accuracy of 61.3%, meaning it correctly classified quality outcomes in approximately six out of ten cases. Overall accuracy represents the proportion of all predictions (both conforming and nonconforming products) that were correct. The model demonstrated a sensitivity of 65.9%, which indicates its ability to correctly identify off-quality products. In other words, the model successfully detected about two-thirds of the actual quality issues.

The area under the ROC curve (AUC) was 0.69. The AUC score ranges from 0.5 (no better than random guessing) to 1.0 (perfect prediction), with 0.7 generally considered the threshold for acceptable discriminatory power. Our model’s AUC of 0.69 approaches this acceptability threshold, demonstrating that it performs meaningfully better than random guessing.

These results suggest that the model can serve as a useful indicator for assessing the impact of product variety-related parameters on quality conformance, albeit with moderate predictive power. While there is room for improvement, these findings provide valuable insights, especially considering the complex nature of continuous process manufacturing and the exploratory nature of this study. The model’s performance indicates that the selected variables do have a noticeable relationship with quality outcomes, though other factors not captured in the current model likely play a role as well.

4.2. The Impact of Each Product Variety-Related Factor

Table 3. Results of the logistic regression analysis.

Variable	Category	β Coefficient	Standard Error	p-Value	Odds Ratio	Lower Bound (95% CI)	Upper Bound (95% CI)
Intercept	-	−1.755	0.050	<0.0001	0.173	0.157	0.191
X1	-	1.078	0.154	<0.0001	2.94	2.17	3.97
X2	-	1.416	0.116	<0.0001	4.12	3.28	5.17
X3	Ref: Long	-	-	-	-	-	-
	Moderate	0.747	0.061	<0.0001	2.11	1.87	2.37
	Short	1.169	0.055	<0.0001	3.22	2.89	3.59
X4	Ref: High	-	-	-	-	-	-
	Medium	−1.378	0.073	<0.0001	0.252	0.218	0.29
	Low	−0.705	0.062	<0.0001	0.494	0.437	0.558
X5	Ref: Very frequent	-	-	-	-	-	-
	Frequent	−0.665	0.060	<0.0001	0.514	0.457	0.578
	Occasional	0.296	0.063	0.622	1.03	0.912	1.17
	Rare	−0.290	0.089	0.001	0.748	0.627	0.89
	Very rare	0.399	0.080	<0.0001	1.49	1.27	1.75

shows the logistic regression results of the link between product quality and the variety-related engineering parameters. A brief explanation of the meaning of the various columns is given below.

• The first column reports the name of the independent variable, while the second column reports the categorical value of the independent variable.
• The third column (β coefficient) represents the change in the log-odds of the dependent variable when the independent variable changes by one unit [37]. A positive β coefficient indicates that an increase in the independent variable raises the log-odds of the dependent variable.
• The standard error reflects the uncertainty associated with the β coefficient, while the p-value indicates the likelihood that the observed effect occurred by chance [37].
• The odds ratio indicates the predicted change in the odds of the dependent variable (quality) resulting from a one-unit increase in the corresponding independent variable. An odds ratio below 1 indicates a decrease in the odds (less likely), while an odds ratio greater than 1 indicates an increase in the odds (more likely) [39].
• The last two columns represent the lower and upper bounds of the confidence interval (CI) for the odds ratio.

Before interpreting the results reported in Table 3, it is useful to underline that variables X1 and X2 are binary, and their odds ratios can be interpreted directly. In contrast, variables X3, X4, and X5 are categorical and should be interpreted relative to their respective reference categories. For example, the effect of moderate uptime (X3) should be understood in comparison to the baseline category, which is long uptime. Notably, X6 does not appear in Table 3 because it was excluded from the model due to low statistical significance.

Moving to the interpretation of the results, we see that the logistic regression model identified several statistically significant predictors of product quality: (1) ramp-up, (2) changeover, (3) uptime, (4) production intensity, and (5) production frequency.

Ramp-up and changeover. First, the ramp-up with minor changeover (X1) shows a strong positive correlation with the probability of product quality deficiencies. Products manufactured during a ramp-up before a minor changeover are nearly three times more likely to exhibit quality issues than those not produced during such conditions. Similarly, products manufactured during a ramp-up before a major changeover (X2) are associated with a 4.1-fold increase in the odds of quality failure.

Uptime. Uptime (X3) was included as a categorical variable with the three levels, namely, short, moderate, and long, as seen in Table 3. The long category served as the reference group, meaning that the reported effects for short and moderate uptime are interpreted in relation to production runs with long uptime. The analysis shows that production runs with shorter uptimes are associated with an increased likelihood of quality nonconformance. Products produced in production runs with moderate uptimes were 2.1 times more likely to result in a quality issue, while those produced under short uptime conditions were associated with a 3.2-fold increase in the odds of off-quality relative to those produced under long uptime conditions.

Production intensity. Production intensity (X4) was included as a categorical variable with three levels ranging from low to high. The latter category was chosen as the reference group, meaning that the reported effects for medium and low production intensity are interpreted in relation to the products with a high number of production orders. Contrary to the focused relationship, the analysis shows that lower production intensity is associated with a reduced likelihood of quality deficiencies. Products with a medium number of production orders have an odds ratio of 0.252, indicating that the odds of quality nonconformance for medium-intensity products are about 75% lower than for high-intensity products (1 − 0.252 = 0.748). Likewise, low-intensity products exhibit an odds ratio of 0.494, reflecting that the odds of quality issues are approximately 50% lower than those of high-intensity products.

Production frequency. Lastly, production frequency (X5) was included as a categorical variable with five levels ranging from very rare to very frequent, with the latter acting as the reference. Although there is no linear relationship, the analysis shows that products that are infrequently produced on the multi-purpose line are more likely to suffer from quality deficiencies than those produced regularly. Production instances where X5 was classified as frequent—meaning that between 31 and 90 days had elapsed since a product from the family was last produced—showed a decrease in the odds of quality nonconformance (odds ratio ≈ 0.514) compared to the reference category. Although not statistically significant, occasional production (91–180 days) was associated with a slight (1.03-fold) increase in the likelihood of quality issues. Rare production (181–365 days) was associated with a reduction in the odds of quality issues (odds ratio ≈ 0.748), meaning that the likelihood of a quality issue was approximately 1.34 times lower than for very frequently produced products. Finally, very rare production (>365 days) had a 1.49-fold increase in the odds of quality nonconformance compared to the very frequent reference category.

5. Discussion

This section discusses the results presented in Section 4, addressing the target research question (which product variety-related factors have the strongest influence on product quality conformance?). The study was motivated by the widely accepted but empirically limited notion of product variety’s negative impact on quality conformance. This narrow focus limits the transferability of findings to other contexts, such as continuous process manufacturing, where the nature of product variety and the sources of quality risk differ substantially. The present article addresses this gap by investigating the impact of product variety on quality conformance in a continuous process manufacturing setting.

Data were collected from a multi-purpose production line used to manufacture chemical goods to investigate the relationship between product variety and quality conformance in a continuous process manufacturing setting. In accordance with the investigation focus, product quality was selected as the dependent variable, while the five statistically significant engineering parameters of X1 (ramp-up with minor changeover), X2 (ramp-up with major changeover), X3 (production uptime), X4 (production intensity), and X5 (production frequency) were included as independent variables. Finally, LRA was selected to determine the relationship between the dependent and independent variables.

5.1. Summary of Principal Findings

Using data from 34,875 big bags representing 56 product variants produced on a continuous chemical line, the findings of this study demonstrate that several product variety-related engineering parameters are associated with a significantly increased likelihood of quality nonconformance. As the level of product variety managed by the production unit increases, the number of changeovers and associated ramp-up periods is also likely to rise [40]. Increased variety often leads to shorter production uptimes and longer intervals between the production of the same product. As the empirical results reflect, these conditions are associated with a higher likelihood of quality nonconformance. The analysis further examined the relative strength of the distinct independent variables to address the research question. Among these, ramp-up with major changeover, production uptime, and ramp-up with minor changeover stood out as the most influential, with each showing a substantial increase in the likelihood of quality nonconformance. Moreover, production frequency emerged as a significant determinant for product quality conformance, with very rare production (>365 days) being especially problematic. Contrary to experience-based intuitions, product intensity was shown to have an inverse relationship with quality issues. Although it may seem intuitive that producing a product more often (high intensity) would lead to greater familiarity and fewer quality issues, the results do not support this expectation. Descriptive inspection of the data indicated that higher production intensity generally coincided with shorter production runs, suggesting that frequently produced products were manufactured in smaller production sequences. To investigate this further, a post hoc analysis of the interaction between uptime and intensity was undertaken to assess whether short production runs are particularly problematic under different intensity levels. The analysis showed a significant positive interaction between short uptime and medium intensity (odds ratio approximately 2.5), indicating that the negative effect of short production durations is amplified when products are produced at a medium intensity.

The logistic regression model achieved an overall accuracy of 61.3% and an AUC score of 0.69. While the predictive accuracy is moderate, it is sufficient for the study’s exploratory purpose. The objective was to examine operational, variety-related conditions associated with increased quality risk, rather than to generate precise predictions. Continuous process manufacturing is affected by inherent sources of variability, which naturally limit predictive precision. Therefore, the results should be interpreted as indicators of elevated risk that can inform further investigation and improvement initiatives.

5.2. Comparison with Prior Research

Our findings extend prior research on product variety and quality conformance, which, as seen from

Table 4. Comparative overview of prior studies on product variety and quality conformance.

Source	Industry	Variety Measure	Quality Measure	Reported Relationship
[10]	Automotive	Option variability	Major rework rate and minor repair in assembly	Negative
[11]	Automotive	Composite indexes of fundamental, parts, and peripheral variety	Defects per 100 vehicles	No relationship
[12]	Automotive	Operator choice complexity	Assembly errors	Negative
[14]	Automotive	Number of factory-installed options on a model	Manufacturing recalls	Negative
[15]	Textile	Variety-related engineering parameters	Composite measure reflecting the volume and cost of output not sold at full price.	Negative
This study	Chemical	Variety-related engineering parameters	Binary quality indicator per big bag	Negative

, has predominantly focused on discrete manufacturing and assembly contexts. While [11] found no statistically significant correlation between the number of manufactured product variants and quality defects, the literature generally supports the assertion that excessive variety has negative implications for quality conformance [10,12,14,15]. While our results align with the general direction of the literature’s claims regarding the impact of product variety on quality conformance, they also reveal a more complex relationship than often portrayed. Our findings indicate that the impact of product variety on quality conformance may be contingent on specific operational factors such as changeover complexity, production run length, and production regularity. This more nuanced view extends the existing literature by highlighting the need to consider these specific aspects of product variety, rather than treating variety as a monolithic concept. The difference between the present findings and those of [11] likely reflects both the industrial context and the way product variety was operationalised. Ref. [11] constructed composite indexes capturing three dimensions of variety: fundamental (number of platforms, body styles, and models), parts (engine/transmission combinations, colour variants, parts commonality, suppliers), and peripheral (installed options within and across models). These broad product-level complexity indicators were examined within a discrete automotive assembly environment, where quality is influenced by assembly precision and organisational coordination. In contrast, our study examines continuous process manufacturing, where variety primarily affects quality through process destabilisation. By employing variety-related engineering parameters that capture these operational consequences, we identify significant effects that are not observable through higher-level product-complexity measures.

Before our investigation, ref. [15] remains the only one to explore the relationship outside an automotive assembly context. Similar to the results detailed in Section 4, her findings indicated that major setups derived from excessive product variety in textile manufacturing are associated with a reduction in quality performance. However, textile manufacturing differs fundamentally from the industrial setting examined in this study. Relative to [15] (textiles), we document analogous mechanisms (setups/ramp-ups) in a technology-intensive chemical setting with more complex and costly changeovers, thereby broadening external validity beyond automotive and textiles. The case company provides a relevant and appropriate setting to test the relationship between product variety and quality conformance, due to its organisational complexity and size (approximately 2000 employees), reliance on uninterrupted production, and broad product portfolio, reflecting the operational landscape many companies face within the industry.

5.3. Implications for Scholars

This study offers meaningful implications for scholars. First, it contributes with a unit of analysis tailored to continuous manufacturing processes (the big bag as the economically meaningful decision unit). Second, it provides empirical evidence that product variety measured through variety-related engineering parameters is associated with increased quality nonconformance probabilities in a continuous chemical context. Together, these contributions help bridge the gap between prior research on discrete and assembly manufacturing and the largely overlooked process industries. They also motivate mechanism-focused studies on sequencing, campaign planning, and ramp-up control in high-variety continuous production settings.

5.4. Implications for Practitioners

The results of this study not only highlight the specific conditions under which product variety is associated with quality conformance but also have practical implications. As the findings reflect statistical associations rather than proven causal effects, they are best understood as exploratory insights that can guide future operational experiments and planning initiatives. First, the study findings suggest that opportunities may exist within planning operations to potentially reduce quality issues. For example, practitioners could consider setting minimum run-length targets by product family, consolidating adjacent orders of similar products into longer campaigns, or avoiding fragmentation of orders that would push runs into the “short” category, as these actions may help lower the likelihood of non-conformance.

Second, while the model’s predictive accuracy is moderate, it can nonetheless serve as an indicative screening tool to highlight upcoming orders with an increased predicted probability of non-conformance. Such indications can support prioritisation of preventive actions, for example, through enhanced quality checks, extended planned uptime, or alternative sequencing.

Third, the existing literature provides several means of enabling and managing product variety, including mass customisation, product configuration toolkits, design for variety, product modularity, cellular manufacturing, and postponement strategies [4]. However, growing product variety quantities have made variety reduction imperative for efficient operations [41,42], and product elimination provides a novel angle for managing manufacturing capabilities through the reduction in portfolio and process complexities [43]. Although the current literature points to various quantitative decision variables for the detection of weak products [42], it contains few that aim specifically at identifying elimination candidates when the need arises from operational deficiencies. While several authors have proposed defect and return item rates as appropriate decision variables for product elimination [44,45,46], the findings in this study suggest that such an approach may be overly simplistic. Product quality in continuous process manufacturing appears to be shaped by contextual factors such as production uptime, changeover type, and production frequency. Therefore, when quality issues serve as a trigger for product variety reduction, managers should consider these underlying operational factors to ensure more accurate identification of elimination candidates and avoid discarding products that may only perform poorly under specific production conditions.

6. Limitations and Future Work

The present study focuses on operational and engineering determinants of quality conformance in continuous process manufacturing, adopting the manufacturing-based definition of quality as ‘the degree to which a specific product conforms to a specification’ [18] (p. 26). It does not encompass organisational, behavioural, or strategic factors that may also influence quality outcomes.

The work presented in this study has several limitations. First, the study is based on data from a single chemical manufacturing company. While the case is well-suited for exploring quality conformance in continuous process manufacturing, the findings may be influenced by the company’s unique processes, recipes, and maintenance practices. Future work should include multi-company, multi-line datasets across process industries (chemicals, food, minerals) to examine site-level moderators, allow for cross-case comparisons, and improve the generalisability of findings. In particular, future studies could investigate how plant characteristics such as automation level, recipe complexity, and differences between continuous and campaign-based production modes may moderate the observed relationship between product variety and quality conformance. Furthermore, as the model identifies statistical associations rather than causal effects, the findings should be interpreted within contexts that share similar production conditions and data structures. Extending the analysis to other process manufacturing settings could also help identify additional variety-related engineering parameters that influence quality conformance.

Second, the company relies on established protocols and tolerance levels to assess quality indicators such as dimensions, colour, and chemical properties. These procedures ensure that quality classifications are primarily objective. While borderline cases may occasionally require assessment by quality control personnel, such instances are rare and unlikely to meaningfully affect the results. Nevertheless, minor measurement uncertainty is an inherent aspect of industrial data.

Third, some independent variables (uptime, intensity, frequency) were transformed into categorical variables for the analysis. While this facilitates interpretation, it may overlook finer differences within each category and reduce the level of detail in the analysis. The model’s performance implies that other important factors are likely influencing quality conformance that are not captured by our current set of product variety-related variables. These could include additional aspects of product variety that we have not considered, such as the complexity of product recipes or the degree of commonality in raw materials across products. Future research should explore the relationship through the application of additional variables.

The present study relied on LRA to model the relationship between product variety and quality conformance in continuous process manufacturing. LRA was selected as the most appropriate analytical approach because the study aimed to explore and explain relationships rather than optimise predictive performance. LRA provides interpretable coefficients that quantify how changes in each independent variable affect the odds of quality non-conformance, which aligns with the study’s explanatory purpose and supports managerial interpretation. The binary nature of the dependent variable further supported the use of LRA. While the present study provides novel insights into the relationship, several opportunities remain for methodological extension. First, the model relied on expert judgement and multicollinearity diagnostics (VIF analysis) for variable selection; future research could complement this approach with systematic feature-selection algorithms. Next, while multicollinearity checks and the use of theoretically grounded variable definitions were conducted, a formal sensitivity analysis was not, as the study focused on explanatory relationships rather than predictive optimisation. Future research could systematically explore the sensitivity of results to alternative model specifications. Third, future research could compare modelling techniques by applying ensemble learning methods such as Random Forests or XGBoost to evaluate potential improvements beyond predictive accuracy alone. These methods can capture non-linear relationships and higher-order interactions among variables that may not be detected by traditional regression approaches. They also provide built-in variable importance measures that could help identify the most influential parameters affecting quality conformance. Moreover, as the present dataset does not explicitly capture temporal dependencies, future work could apply time-series or dynamic machine-learning models to account for evolving quality patterns over time. Finally, the inclusion of deep-learning architectures or neural networks may uncover complex, multidimensional relationships among product-variety parameters and quality outcomes that extend beyond the capabilities of traditional regression models.

The findings presented in this study offer important information for defect-reduction policies in practice. Future work could operationalise the findings for decision-making by translating the model into planning tools (campaign sizing, sequencing, elimination screening). To further assess the applicability of the findings in practice, future work could also explore how the identified predictors for quality conformance can be operationalised into product elimination models, allowing companies to target variety reduction efforts based on operational impact. In continuation, longitudinal studies that examine quality performance before and after a variety reduction initiative can provide insights into the impact of such efforts. By tracking quality metrics over time, researchers can assess whether the elimination of specific product variants leads to measurable improvements in quality conformance or whether issues persist due to other underlying operational factors.

7. Concluding Remarks

This study shows that in continuous chemical manufacturing, quality risks associated with product variety appear to be concentrated around conditions such as changeovers/ramp-ups, short run lengths, very long inter-run intervals, and high-order frequency when coupled with short runs. By adopting an exploratory approach, the findings of this study extend the research on product variety and quality conformance beyond discrete manufacturing and assembly contexts. Practically, the findings provide initial guidance for production planners and portfolio managers seeking to balance product variety with quality conformance in process industries. Overall, this study highlights the need to consider product variety not only as a strategic product portfolio concern but also as an operational factor that may influence quality performance.